 Good morning. Good afternoon. Good evening to all. I would like to welcome you all to this bow in food seminar on food matching. We have great presenters in front of us and covering practical and theoretical issues on food matching. It is a very interesting and important topic, not only for food composition, but also for dietary assessment because the quality of the food matching will determine the quality of the outcome. It is a food, a nutrient profile, or it will be the nutrient intake and everything which comes beyond that, which can be some decisions for a program or a policy. So I would really like to welcome the presenters. So we will have Fernanda Grande who will give us a presentation on the Bow In Food guideline for food matching. Judith will talk about, Judith Kanmingen will talk about the food matching approach used in the Caribbean for 30,000 foods. Alpressa will present the computerized food matching approach and Bridget Holmes will talk about the FoodEx2 facilitating food matching. For your information, you as the attendees, you will be able to ask your questions. There is a section below on Q&A or chat. This is where you can put your questions and we will all answer them in the end. So I would also like to thank all the presenters who have accepted this invitation and I'm really looking forward to a very nice seminar. Fernanda, you have the floor. Thank you Ruth. Let me share the screen and the presentation. Well, thank you for this introduction. Let me just change here. Now I can also see the presentation. Hello everyone. My name is Fernanda Grande and I'm talking to you today from São Paulo in Brazil. And I have been working as a food composition consultant for FAO since 2015. And first of all, I want to thank Ruth for this opportunity to present on another very interesting FAO In Food webinar and today I will talk about the FAO In Food guidelines for food matching. So in the next 20 minutes I will cover the definition of food matching and some key challenges. And I will talk about the FAO In Food guidelines. So starting with the definition of food matching. So as Ruth already mentioned, when we say food matching, we are referring to the activity that links food from different data sets. So it can be linking consumption or supply data with food composition data to allow for nutrient intake estimates or linking food from two different data sets. And also a very important step during the compilation of data, food composition databases, both during the data aggregation and also for borrowing values to fill the gap from other food composition tables. When we are performing the food matching, we need to keep in mind that it is extremely important to ensure that the food from the different data sets are as similar as possible. So meaning that the food composition data from one data set corresponds to the food that we want to represent from the other data set. This is why food matching procedures are critical to obtaining both high quality estimation of nutrient intake and also food composition databases, since most of them combine and borrow data from different data sources. But as food consumption and food composition databases describe and classify data using different criteria, it makes of these activities difficult and challenging tasks. And I brought here just a couple of examples, so food may be not very precisely described in the composition surveys, especially in household budget surveys. But of course, that the details included in the food description will vary a lot among the different surveys depending on their characteristics and purpose. So having that said, let's suppose that we have this food in a food consumption survey, so being unknown type, try it and boil, and that we need to match this food with another one from a food composition table. So I took CFAO in global food composition database for pulses, and I selected all the foods that could match this food that is described in our hypothetical survey. So I could find 12 different types of dried or major beans, in this case they mean the same, that were also boiled. And we can see that in the food composition table there are additional descriptions, so we can see that all the beans were water soaked and boiled in different water. And additionally in this data set we have two options of cooking methods for each of these beans. So we have an option that is boiled and then the water is drained, and we have the option that was boiled and the cooking water was not discarded. So therefore in this case we need to decide how to match these foods. So should we take an average of all beans in the food composition database, or should we make an average of the known as the most consumer ones using also the most common cooking practice for this given population. So these are just some options that we have here. On the other hand here we have a food with more details in the food description, so we have lentils, brown, dried, it's filled with butter. And when looking at the same database again, these are the two foods that I could identify as the most similar ones. And as you can see, we have differences in the food description. So what is more important here is that the details that are different among the food descriptions, they affect the composition of food. So for example, with butter means that the fat content will be higher. And if we use either split or whole lentils, it will also change the dietary fiber content. So in both cases that I showed here, we couldn't find an exact match for the food that we are looking for, so we need to take a decision about the best procedure to match the different food. And this decision can change from food to food and also depends on the information available, but it's important that some rules are set to allow for a standardized food matching during this whole process. So during the food matching, it is important to keep in mind two key aspects. So first, the food matching should always aim for the highest quality match possible. And usually it's done by identifying the most appropriate food in the most appropriate source of food composition data. And also to properly access the nutrients in case there can be no missing values in the food composition data. So it might seem simple, so basically the food needs to be equivalent and we need the complete nutrient profile. But anybody who has already worked in this task, we will agree that in fact, this may be a quite challenging activity and that it's not always possible to achieve a high quality match. And this is why FAO and Infoot have developed the guidelines for food matching and these guidelines were developed with the contribution of a group of experts that I listed here. So the guidelines were created to assist in selecting the most appropriate food to match to the food reported in food consumption surveys, but they are also very useful when compiling food composition databases. And to develop this guideline, a call of experts was issued using the Infoot mailing list and the experts were guided by a FAO coordinator and several versions of the guideline were reviewed and discussed until having the final document. As a result, the guideline is divided into main parts and in the first part, the guidelines include the general aspects on food matching, which lists some key aspects that we need to take into consideration before starting the matching process. And I will briefly talk about each of these. So first we need to identify the food component of interest in some decisions on the food matching depending on the components of the survey. So for example, if the survey is interested in vitamin A intake, it is of great importance that all the factors that affect the composition of vitamin A in food are taken into consideration. On the other hand, if the survey does not look, for example, at sodium content, it doesn't matter if the foods are matched with a salted or unsalted version. And it's also important of course to consider the components that are presented in the food composition table that we are going to use. Since if we have a gap on the data, we will need to look for another matching in another food composition table. Second, we need to identify the amount of food consumed because this will also guide us on the priority food for matching. So since the food matching is a very time consuming activity, especially when we do it manually, which is my case, sometimes we need to decide on the priority food to direct our efforts to the most relevant food. So the food that are consumed in high amounts or with high concentrations of the components of interest should be considered as a priority and we should aim for really high quality food matching. But if we have foods that are consumed with less frequency and also with low amounts of the components of interest, we may use a lower quality food match in these cases. Third, for foods that are not clearly described in the survey, it is important to identify the eating habits of the population to attribute a best food match. And in some cases, it is important to identify several foods and take an average to better represent the foods that are consumed in the country. And fourth, it is important to document the matching quality. So assigning quality codes are a very useful procedure to access the quality of the food matching and consequently of the nutrient intake estimation. And the guidelines also list an example of quality code. So it suggests that we should attribute codes starting from the high quality match and where we have exactly all the food description and all the descriptors are exactly what we want. And also the food component represents the components that we want to use until we we go to the lowest quality match where the matching is just the closest match as possible. So the number of subclasses of these quality codes may vary according to our needs. So we can set, based on these examples, we can set our own subclasses for classifying the quality of the food matching. And the second part of the guidelines, it brings the food matching criteria itself. So it is important that we should consider not only the food identification, but also the identification of the component. And here we have some specific criteria also. And let's start with the food identification. So first, we need to consider the food name, of course, and the food descriptors. And ideally, they should be complete and unambiguous in both data sets, so in the food consumption survey and also in the food composition database. So we are sure that we are taking a similar food. And it's also important to consider that the food name can vary among countries. So sometimes we can find things like groundnuts and peanuts, maize or corn, and they refer to the same food. They are just different names to refer to the same food. So it's important to take this into consideration, especially when using food composition tables from other countries. And the other criteria that we can use are the scientist names, but we also need here to keep in mind that they can vary also for the same food. So I'm bringing here an example for broad beans, we can find all these scientist names that I am bringing here. The first one involved is the official one, but we can also find the others. So there are some websites where we can find the scientist names available for a given food. And I'm leaving here this too, just as a suggestion if you need to search for this information. And the last criteria, it is important to check also the water and fat content, as well as other components of interest. So water we need to check for all foods to see if they match, if the water content match the food description, while fat is important for foods where they vary a lot, such as meat, milk or dairy products, and other components of course should be checked depending on the components of interest in our calculation. So regarding the food identification, the guidelines also include a list of some important descriptors that we need to consider. So they include the processing and preparation state of the food, color, biodiversity, material, stage part or source of the food, refuse an edible portion, and fortification for processed food. And for each of these items that I'm presenting here, the guidelines also bring some important questions that should be asked. So just to show you how it is described in the guidelines. So here regarding the processing and preparation state of the food, we can find the questions to ask and things that we should watch for doing the food matching process. And we have this for all that descriptors that I have presented before. So just to illustrate the importance of a complete food description, I'm bringing some practical examples. So if we have a food described as cassava raw, we can see that in the food composition table. We can find the cassava tuber and also the cassava leaves. So the omission of the part of the food in the food description in the survey in this example can lead to very different food to match in the food composition table. In the second example, we have tuna canned and drained. And in this case, what is missing is the preservation mean. And when we look into the same food composition table, it could be canned in water or canned in oil, and it will of course affect the fat content, but also the micronutrient content for the given food. And my last example here, we have sweet potato flash raw. And in this case, what is missing is the color of the flash and the color is related to the vitamin A content, meaning that it would be of great importance if vitamin A is one of the components of interest in our survey. So as you can see here, the content, the vitamin A content will change a lot depending on the color of the flash. So this is why it is extremely important to have a complete food description in both data sources that we are matching. In addition, the guidelines also brings in the final part on the next one, some examples of challenges that we can face during food matching, along with possible solutions. And it can serve also as a guide for us to take decisions during the food matching. So I really recommend that you look at this part, if you are performing the food matching for any purpose. So the next key aspect that we need to watch for when when matching food is the food component. So usually we look a lot on the food identification, but the identification of the component is also of great importance. And it includes the component identification and the units and denominations. So why we need to check the component identification. Some components, they have the same common name, they can be presented in the food composition table with the same name, but they may represent, they can represent different components for making sure that we are using exactly what we need. We need to check the expression definition and analytical methods. So just to give an example, regarding the analytical methods, we can have either crude fiber or total dietary fiber and we know that the values are completely different. For the definition we have, for example, vitamin A expressed in retinal equivalents or retinal QT equivalents, and they change a lot, especially of course for plant-based foods. So thinking about the components that we need to be really careful, we have energy, protein, carbohydrate, fat, fiber, and the vitamins A, D, E, C, folate, B6, and niacin. So for all these components, it is extremely important to really check the component identification when we are using more than one food composition table. And also the units and denominators. So the units are usually gram, milligram, or microgram, and sometimes kilojewels and kilocalories for energy. And the denominators that are, that represent which amount of the food we can find the component that is presented in the food composition table. So usually they are presented per 100 grams of edible portion, but sometimes they can be expressed also per ml, especially for liquid. So we need to use the density conversion factors in this case. And the guidelines bring the dynamics, too, at least of the most important tag names. The tag names are the input components identifiers, and they are really useful for matching the components. Especially when we are using different food composition table, matching the tag names with the components from each food composition table will help us a lot to avoid inappropriate mixture of the components. And finally, just to conclude, I'm listening here how I have been using the guidelines for matching in my work experience. So of course I have used it for one minute left. Okay, from food consumption survey and purchase data with food composition table, and it is extremely important and useful for tracking rules to make decisions using standardized approach, especially when we are working with a group of people doing the matching rules so that we can follow the same rules. And also, when we are matching food from an unknown country for us, when we don't know very well eating habits, the guidelines help us to make the right questions to check the matching. And also, of course I have been using a lot for the compilation of food composition table. It's very useful for matching food during the data aggregation and also get feelings, but it also helps us to make a complete food description since it helps us to think about the descriptors that change that affects the composition of food. And also, to decide the food entries that we want to present based on what affects the composition of the given food. So as you could see the guidelines present a valuable, valuable information for food matching, and I want to take this opportunity to thank FAO Infos and all the authors that contributed to this important guideline. And if you don't have the guideline yet, you can download it directly from the FAO Infos website. And that's the end of my presentation. Thank you all for following it. And I will be here for the questions at the end, but if you wish, I'm leaving here my email and also my website Food Composition Explained where you can find some more useful information about food composition. And thank you again. Thank you so much, Pananda, that was a brilliant presentation on the overview and in addition, you gave some very good information on food description, component description, so really thank you so much. It will be very useful as well for the next presentation and we are looking forward to Judy's presentation on the food matching approach used in the Caribbean for 30,000. Good evening everyone. Okay, good evening everyone. And thank you for participating this evening and FAO for giving us this opportunity to do so. I'm Judy Cunningham and I'm in Australia and I've had a very long career working in food composition and national nutrition surveys. So I'm just going to share my screen. And tonight, because it's a nighttime here, excuse me, going the wrong way. I'm going to talk about a project that I was involved in with Fernanda and Ruth last year to match food, extensive food list of foods reported in a national nutrition survey in the Caribbean to nutrient data. So a bit of background to the project. There's been a national nutrition survey being held in St Kitts and Nevis and St Vincent and the Grenadines last year and continuing this year. It's been managed by the University of the West Indies and it's had a lot of support and assistance from FAO. In mid last year, the university engaged three consultants to lead the food matching component. So that's the component that takes the food list that are used in the nutrition survey software and matches it to nutrient data for those foods. And the team of three consultants we had was Fernanda who's just spoken to you. And Dr Daniella Beltram, who is also from Brazil, but based in Paris, and myself and I'm based in Canberra in Australia. On our food matching team, we had four students who were gave us invaluable assistance, Chaneli Bryce and Catherine cargo Warner in Trinidad, and Taiz Rodriguez and Glenda Vian in Brazil. Our part of the project lasted for five months, but the survey work is only recently finished. So what did we achieve? We matched over 30,000 foods in the food list to nutrient data. And for each of those foods, we had up to 38 components, as well as density, edible portion and a quality rating. So in all, we produced more than one million data points. And if you think about one spreadsheet, how big that must be, you can see that it was actually quite an achievement to get this work done. There were very many challenges that I will talk about a bit more, but I must talk about the two most important ones at the beginning. The first was time we only had five months to match over 30,000 foods, as well as carrying out recipe calculation and preparing training materials. I've been involved in a couple of nutrition surveys in Australia and we spent much longer than that to do our food matching. So obviously we had to fit our project into that timeline. Of course, the other problem was everyone's friend COVID-19. This meant that there was no travel, no face-to-face meetings, very limited ability for people to go out and check in retail outlets, in markets, and people working from home. We couldn't always access libraries and documents. So this added another level of complexity that I hope that future projects don't have. Some of our other challenges included that the food matching team became involved late in the project. And so we didn't really understand how the software worked for the survey. We had some uncertainty about doing our food matching. The team was spread across the world from myself in Australia through to Ruth in Chile. So we had a very large time difference and that made it very challenging to find meeting times. And people were very flexible about meeting at dinner time or very early in the morning. The three of us consultants were not very familiar with Caribbean food. So that presented some challenges in identifying foods. And also for some of the local foods, we just could not find existing food composition data. So how did we carry out the project? Well, overall we can summarize it as doing some background research, agreeing timelines, preparing project plans. Then we established a reference data set and a file format for data exchange. The main part of the project was actually carrying out the matching and we were very much guided by the guidelines that Fernanda has just talked about. There was a phase of checking the data and finally publishing the final files and report. This one, this screen is just to give you an example of one of the planning tools that we used because this was a very large project with many people involved. It was really important to develop a simple timeline that people could see where we were in the project and whether we were going to meet our timelines. And this project, this Gantt chart is available free of charge in Microsoft Excel. So there's some very easy tools to use to help. Some of the background research that we did, we looked at what are the major staple foods of the region because as Fernanda said, these are the ones for which you want the best quality matches. We wanted to know about the cultural and ethnic heritage of these two countries because that will guide us in selecting matches and understanding recipe dishes. Where do people in the region obtain their foods? Is it from supermarkets where foods might be fairly standardized? Or are they from street stalls or growing in their own yards? Are foods produced in the region or imported? Where are they imported from and what are the laws that govern the composition of food? And a big challenge we had was how to deal with local foods. So these could be foods such as ones that were formulated to meet specific regulations in those countries. So, you know, whether iodine had to be added to salt or how much alcohol could be in a beverage. Or they use local ingredients such as the local fish and fruits and these were very difficult to find data for. Or they use unique regional recipes. One of the things we decided very early on was which components to include in the data set. We aimed to include all the nutrients for which dietary reference intakes exist. And in this region they rely on the United States dietary reference intakes. We were also interested in components of current or emerging public health concern, particularly sugars and ethanol content. But we only wanted to use reliable comprehensive data for those components and we had wanted to include data for vitamin D. But we felt that we didn't have enough data to allow us to do that so we didn't include vitamin D. We also needed data for density and edible portion. Now the components to be included were agreed by the full team at the beginning of the planning stage. Other things that we agreed was that the data is for the purpose of estimating nutrient intakes. It's not data for producing a reference food composition table. And we couldn't aim to make a food composition reference table because we had such limited time and such limited staffing that we could never get enough data of sufficient quality. For example, we had no time to go and search literature to find data for local foods. Nevertheless, we agreed that it was very important to document the data origin and any assumptions that we made any adjustments we made. And that we should assign a quality rating to each match as Fernanda has outlined. One of the most important things that we did early on and I thank Fernanda for doing this for us was to prepare a consolidated data set that we would use for our matching. So the data set included all the components that we wanted with the tag names that we agreed we would use when we couldn't get data with the tag name we wanted. For example, data may have been carbohydrate by difference rather than available carbohydrate. We agreed to enclose those values in square brackets. And we also agreed to standardize some values so that we had, for example, the same method of calculating energy for all the foods in the data set. What we ended up with was a data set of more than 32,000 foods that was searchable and shareable and easy to match to the foods in our food list. We use Microsoft Excel because all the team members had access to this and knew how to use it. We did not have time to set up a database system or to train people and how to use it. And we arranged the data in the way we wanted to arrange it in our final file format. So we weren't having to rearrange columns because that's very time consuming. This is just an example of what our reference data set looked like. So we had a source code that we generated, the name of the country, the name of the food, the ID code in the national data set that it came from. And then we had all the different components heading away in 38 different columns with some approximate energy water. And that, as I mentioned, had 32,000 entries in it. So what databases did we use to make that reference data set? Well, as I mentioned, we didn't have the time or ability to do literature searchings or to go and survey the composition of local foods. So we had to rely on data sets that were available electronically, are free of charge and contain high quality data. And we did find out early in the project that most of the food in the region is imported from the United States. So we decided to use US nutrient data as being the most relevant to this population. And actually many foods on the food list were taken from the most recent US food and nutrient database for dietary studies. So we also use that. And because the ethnic background of many people in the region is from Africa and India, we also drew on the Indian food composition tables and the tables for West Africa. And also because some foods come from the United Kingdom, we used McCants and Whitteson's data set as well. We also drew on some other national food composition tables, FAO publications for pulses and fish and shellfish, and some online resources such as the Q gardens botanical database and also company website. We agreed the file format at the beginning as I mentioned so that we could easily paste data in and that also was a format that could be used to upload data into the survey system. So okay, I've covered that one. So this is an example of the file that we came up with at the end of the day. So if you see these columns starting with pink, these are the food list columns that we were given to work with. They have a food group, a food name and a unique identification code that was the way that data is loaded into the software. So we just reproduced that in our final files. We added two columns at the beginning about the quality of the data. Our quality code was assigned for every food. And where we got data from another source and I'll talk about this later, we put a comment about it. So in this case with rice flakes, the retinol value is imputed as zero. And then we included the code of the food we matched to the country it came from the name of the food and pasted in all those nutrient values. Once we got used to using it was a very easy way to do the matching. However, matching is very time consuming. Originally we estimated we could do no more than one minute per food. So we had to find ways to speed up the process. We focused on major differences between foods. We might have matched to a similar food instead of an exact match. We made some assumptions about sodium reduction for voluntarily fortified foods. We focused on the most important, the most common type of fortification. And we used a not further specified fat line for all our fried foods rather than looking at individual with butter with olive oil with coconut oil. We had some real problems with the matching. And I mentioned before we didn't have a very good understanding of how the survey software deals with things such as addition of salt, sugar and source. So we probably should have clarified that early on. We were unclear about food fortification because of COVID we couldn't get in touch with the people we needed to speak to about this. In the end we had to assume that fortification in the US is the same as in the Caribbean. And because we couldn't go to the local shops, we had to use online resources such as supermarket websites, even YouTube channels for looking at how people make recipes. I'm going to give you a few examples of some of the foods we had challenges with. The first is a food called roti dalpuri, which I had never heard of. And it appears to be a mixture of a flatbread with some meat in the middle. So first of all, we had to do a Google search. What is this food? And then we had to find some pictures of them to see what they look like. We found some local recipes for them on YouTube. And then we found the closest match we could to the main components and estimated the proportions of the main components. Clearly you can see that that process is not good enough for a reference data set, but it was okay for a national survey. So particular types of rum that are drunk in the Caribbean, very high alcohol content. And they're very popular. So we had to calculate a new nutrient line with a higher alcohol content and a higher energy content. Ackee fruit. We had very many lines for this local fruit. We didn't have time to calculate nutrient values in the cooked version. So we just matched to raw. We didn't know what the edible portion was. There was nothing anywhere we could find. So one of the team in Trinidad picked some local fruit and weighed them in her home. And then we had these three, this tuba, which in the food list was called Edo's, Dasheen and Tanya. In fact, these are all types of taro. So we were able to match to taro and that saved a lot of time. We just matched them all identically. Fernanda mentioned briefly that you often have a situation where nutrient values are missing in your data set. And this was something that we found. And we don't want to have missing values if we're trying to estimate population intakes. So we tried to fill these missing values. It wasn't always possible because this is a very time consuming process. But we aim to get 90 or between 90 and 100% data filling. We always recorded the source of the estimated value or the imputed value. This is just an example of what it looks like. So for example with cheese, you see we have some notes here. Fiber and alcohol were imputed zero. Folate FD was imputed from total folates and we assumed that folic acid was zero. So you can see we've got some values that are highlighted in red. That was our way of showing that this was a value that was not from the main record. And where we couldn't find a value that we were confident in, we just left it blank. At the end of the process, we had to check the database. At any time you produce a large database, you must check the outputs. We started the checking very early in the matching process and it continued right through to the very last day when we sent in our final files. If you don't start early in the process, you'll run out of time and your eyes will be falling out of your head and you'll be falling off your chair with checking. So it's very important to start early. All the matching done by our students was checked by a consultant. But I think in any extensive data set, you will have errors in the data. And I think that is just a fact of life, particularly when you're working on the very tight deadlines. After all this work, we produced our final database. So what were the things that helped us to get through this project and get it completed on time. I think that we had a very strong team with a range of skills and everyone on the team was very flexible in how they worked and we were able to share tasks. If one of us had no time that week to do something, then one of the others would do it for them. And the team worked very well together. And this is a really important factor in the success of any project. The consultants all had a background in food composition and experience in preparing data sets and in managing large projects. The university set up a Google Docs page where we could all access it from all around the world and we put all our information there. That was very helpful. So we didn't have missing data where someone had forgotten to send something vital. Most of the time we all had access to a good internet service, not always when people were working from home. And the support from our students who were helping us was really important because they did a lot of the very routine work for us. So what recommendations for similar projects in future involve food composition data matching at the beginning of the project, not as a separate project that's unrelated. Assemble a team with a good mixture of skills. Keep your food list as small as possible without losing the detail about foods that you need to deal with the nutrients you want to report. 30,000 foods was a very, very large data set to work with. And if we'd had fewer foods on the food list, we could have spent more time making quality matches and researching literature. Give careful consideration to how the foods are named because this really helps people to search easily and find them. And we found that foods that were taken from the US survey data set, sometimes the names were changed. Maybe there was a spelling difference or the word order changed, which meant that you couldn't do an exact match search and that slows down your searching. So I'll be very interested to hear from Carl and Bridget about automating that process. As I mentioned earlier, it's really important to do some project planning at the beginning so that everyone knows what they have to do and when they have to do it. And make sure you involve younger scientists so that these skills transfer across generations. At the end of my talk, once again, thank you FAO for giving me this chance to talk and thank you to the team that helped us in the project. Okay. Well, thank you so much Judy for this very wonderful presentation on on everything, the challenges as well as the success of this very, very, very challenging exercise that you have done with the team. And I think it really came across very nicely, what you need to do and what you need to avoid and what are the factors of success. So, and yes, it was a huge amount and the idea was behind that we will follow as much as possible the ethics of philosophy that if somebody does not know an exact food that they would not be forced to, to indicate the most closest and so it is better to do it afterwards than to do it while interview. This is why the food list was so huge. But sure enough, it was a challenge and you manage it and congratulation to you and the entire team for doing so. I probably one comment on that one as well. When we got the data back, a lot of people said that they did not know part of the food description so they use a lot. I don't know. So, meaning that our approach was the right one. But well, this is something different. Now I would like to give the floor to Carl Presser to talk about a computerized food matching system, so which might make things so much easier in the future and let's see how it could work. Thank you, Ruth. Let me start in full screen. First of all, I also would like to thank FIO, especially Ruth, to give me the floor to present what we have done so far. So, my name is Carl Presser, I'm based here in Switzerland. My background is computer science. And so you realize then that's why I'm okay we started a tool, try to do this matching computerized and let's say automatically but also halfway automatically because the expert is always needed. So I can bring as a spoiler without expert matching is almost impossible. I would like to present here a work that we did together with the Swiss government and university, DJ Zurich, in this matching. So let me start with a challenge. So we had in Switzerland a food consumption survey, and we had a pilot study where we first tried the whole process, doing the interviews and then try to match with the food composition data. But just to have a test run with about 5000 unique foods. And then we went over to the main study, where we had almost 36,000 foods, unique ones. So, basically it means 130,000 food items were collected in interviews. And we are then aggregated this food items so where when they're all were described in the same way, and they were for us, the same food. And so we are condensed them to 36,000 unique food items and so you see kind of the first step tried to bring to get lower the number. On the other hand, we had the Swiss food composition database with about 11,000 foods. And the task was as the webinar is we wanted to match this data sets that we can make an intake analysis. So, to show also to start with a practical example and we had on one side kind of, we were looking for a potato so we had to have kind of country fries or they were cooked potatoes with the peel. We had potatoes on your keys you see also kind of just specify a little bit more the other food. So just had potatoes on one side. And then the other data set we had potatoes chips. We had potatoes peeled and steamed potatoes peeled and cooked, or potato peeled and raw and now the question was okay so probably this foods are a good match. But what about with just potatoes so as we hear from several times before them. Yeah, so that food description is quite important them if you want to have a high quality matching. So for system the full case system let me quickly explain that you understand the background so we have a system Switzerland. And in several other countries using food case for managing their national food composition data so they're collecting data and put it into the system they calculate recipes aggregate data. And the system can also manage the food consumption data from from consumption surveys. We also extended the tool with to manage total diet study where the focus is more and more on the harmful substances not on the beneficial ones. And finally the latest module is branded food that you also can store branded foods in the database which are very similar to food composition but not completely the same. So this is an overview. This is a bigger system where we have a background of the database you have some clients also desktop application where you can manage all your data and some dissemination part that you have a website and also with a mobile app, which is kind of for helping in the TTS. So, how was the data flow so our full consumption survey we did with a pick soft nowadays the change the name to global diet. So, this was the interview data we needed to import. We also have in the system some some flexible excel import features where you can just import the excel files, or always possible that you manually enter basically data right so the point is so we imported all the full composition we had in the system already. And then we run through a six step matching wizard we call it, which basically at the end you had all the matching. And finally, once you have matched the full composition data set with the consumption data set, we have an export wizard, which allows you to define what data basically want to export and then an excel file is produced for for the further analysis. So, let me show the first concept in this matching wizard. So, the first concept is a matching run. So basically we say a matching run. It has a from data set and the two data set and a from data set is for instance the consumption survey from 2020 and the two consumption sets so we want to match to this full composition data set from 2019. And you see another run is for instance we take the full composition from 2019. So we take first the full composition site, and we want to match it to the consumption survey from 2020. And the third run a third example is if you take the full consumption survey from 2020 again, but this time you match it with the full composition from the year 2021. And so you see runs is something that's matching this is an ongoing process so maybe you have a new composition database and want to match it or you have a new consumption database. And sometimes you also maybe use the composition with brands and sometimes maybe only the generic food so you can also play a little bit around and check what's the outcome. And that's why that is concept of runs so that you're more flexible to match. You see also you're an example in the food consumption side if you probably have let's say six food just to simplify it, while on the food composition you have only two foods. And now you start to matching as so for all of the six on the left side of front side, you matching all of the six maybe for one you just don't have a good match you just skip it. But all the others you have to match the two, which means from the full composition side you have several matches from this side so that's why also that we have the from two data sets you can change, plus having multiple multiple runs. Then the matching the automated matching, we were investigating and we came basically this approach, which is, we take the names, we have basically multi step approach, I show you why so the first one is we take the name. So you see on the consumption side, we agree that the food is called fresh apple, while on the food composition, maybe it's just called apple raw. So, is this the same food or no for the machine, it's a little bit harder to decide as for the humans. The algorithm is quite simple, take the first three letters, the FRE, take the second, the third and the fourth letter, put it into the group, take the third, the fourth, the fifth letter put it in here and so on. So you're producing triples of characters, and you do this for both sides. And then the algorithm basically saying, okay, now how many you have in common, for instance, this app, the starting of Apple, you have common at the PPL, you have also in common and so on so that the formalize basically check the intersection, which you have in common. And what is the whole union of it so we have three over seven, which is a similarity of about 43%. As I was saying, okay, this food with 43% could be the same because the name is quite similar. I said, okay, good sounds like a good approach, which is also very common in national language processing, but we quickly identified an issue so if you take apple pie and the apple raw, we have a similarity of 60%. So the algorithm will propose you to match the apple raw with the apple pie, which is of course completely, yeah, not not good matching so we realized okay if you only rely on the food name. This will not be enough for this one. That's why we needed to go further first of all was then including synonyms as we have here for before that the synonyms also should be involved and what we also involved is in the global diet you have distance facet descriptors describing the food with several facets. And this we also need to take into consideration because we had on the food composition we did not have descriptors we just had the name, while on the consumption database. We had all this descriptors. And so we also needed to take this into account and check their and the similarity with including the descriptors. And so we could kind of collect additional points so in each step and we reject the food names with all against each other, and we're giving some some points there. The next step we also realized the food category we have to take into place so if you have on one side food category called vegetables and then the source and example just two of them. On the other side you have food groups with vegetable fresh and sources. You have some foods below this food groups so if we want to match a tomato or for instance on the other side we only have tomato. We are if we would have a matching of the food category so we know vegetable is best fitting to vegetable fresh and sources is best fitting to sources. So we can give basically additional points if the foods are in corresponding food groups and so we could give here additional points for this one and we're gaining basically more precision, this one so here you see kind of a print screen of this matching So we had a pre matching a step where you need to match the food categories from both side, because this both those data set we had completely separate developed so they were not kind of used for one to the other. They were completely independent so we had a first step where we needed this category matching step. The next what we also implemented, but what we couldn't use is the food x to of course food x to I would be very helpful if both sides would use the food x to but we did not have this on both side. So we couldn't use but basically implemented that that we say okay if you have a food in food consumption which is kind of described until level four in food x to while the food composition was describing it until level five. Now we can kind of check the pause right if this food have the same pause in the food x to hierarchy. We can also say that the similarity of 80%. And so these two foods could be. Yeah, very probably the same, because 80% is quite high and we're at level four already which is quite a good match basically. So food x to with much help very much. The new food x to version with all the facets of scripters makes it a little bit more challenging because now you can describe the food with several facets. And there you also then have to evaluate the similarity, which then a little bit more challenging but still possible. What we also did is the same with language because language is also them nicely describing the foods. And so we also could kind of evaluate the similarity between two foods based on their language description. But also here we couldn't use it because the full consumption data we did not have the language description in this case, but we just also kind of added this to the model. The algorithm was calculating all the similarities and came up with suggestions because now, if you have for instance here a measure from the full consumption, the full composition lists you the most probable matchings and gives you a similarity. So you see we have a list which is descending ordered with the similarity. So we have the most probable food on top, and the second most probable on the second place and so on until the end where you say, you can search your own food because the matching was not really good. And so you see the list you can select if you're happy with this one and go on and here is kind where we made that the expert coming to play because the algorithm could only propose something but with the final the expert needed to decide yes this is the right matching or no this is absolutely wrong one number four is much better. I take the entry on number four. So this was very important here to bring the expert back into play because the challenges. It's quite a big challenge. What you did also according to the FAO guideline is we added for each matching you could kind of give a confidence which saying I low medium or good so you're also a little bit rating or assessing the matching. And you was also able to put a comment there so that you say oh I took this food because of certain reasons as we added also this information that you later also have a little bit documented. Confidence also says a little bit maybe about quality so if you think the nutrient is profile is fitting very well to it and then you give a higher rating otherwise you probably give a lower rating for this one. Once you have done all this checking there what the algorithm basically and proposes. We also had this review frame where then people can go back and checking okay how many times a food was mentioned so we see for instance your tap water. We had in one study we had three and a half thousand mentions of this food you see also we condensed because all the descripts were exactly the same. We said okay we only have to match it once and then we can reuse it. And you can also then hear review you can search you can maybe all the low confidence matches you can revise by a second person. And you can also batch process several at once you say all the energy drinks for instance I just match with with an average energy drink. This export tool that the end you can just say okay give me all the data that the fields that I want from the consumption table and all the fields from the full composition table and export it into an Excel file, which were then later used in statistical programs like our SPSS for further than for this nutrient analysis. So I thank you for your attention. That was a very nice presentation as well thank you so much Carl for presenting on how the future might be looking like if we have not only our Excel spreadsheet but we have really a software which will help us to see which one is more consumed than others and then concentrate on these So thank you so much and you mentioned food x2 so how nice that we have now Bridget talking about food x2 facilitating food matching. Bridget to have the floor. Good morning, good afternoon and good evening to all my fellow presenters and all webinar attendees. I just like to start by thanking Ruth very much for the kind invitation to present today, and to also thank you all for attending this webinar. My name is Bridget and I'm a recently appointed nutrition and food systems officer and group leader for the nutrition assessment team at FAO in Rome, and today I'll be presenting food x2 facilitating food matching. So I'd like to like to start by extending a very special thank you to the nutrition assessment team members who work very closely with food x2. The team members include Victoria Rita and the excursion razor and some of these people are attending the webinar today and indeed Victoria will kindly be available to help with technical questions in the chat if needed. And I'd also like to just mention very quickly the other nutrition assessment team members, including Fernanda, who's recently joined our team this week so really delighted that that you're with us Fernanda. So to start with a bit of context and as as I'm sure you're already aware, aware, and there is a general lack of dietary data, especially in low and middle income countries, and it's the aim of the nutrition assessment team in FAO to accelerate data sharing and maximize data use and create a robust global evidence base on which to develop policies and improve nutrition. And when I refer to data here I'm referring specifically to data on food consumption, food composition and data on dietary adequacy quality and diversity. So data on food consumption and partly food composition that I'll be focusing on today and how food x2 is used to facilitate food matching with these data. I'd like to highlight some of the reasons why we need dietary data before continuing. And I couldn't fit any more reasons on the slide but I just wanted to go through some of these points. So for anybody who isn't already aware about how important dietary data is. So we need it to just understand basically what people are eating and drinking, understand a bit about the eating context so some information perhaps on where people are eating and when they're eating with whom they're eating and information on meal patterns and habits, but also to consider differences in dietary intakes by factors such as sex, age, rural, urban income level family size. We can also use dietary data to provide evidence on energy and nutrient intakes but also understand nutrients in food and the sources of nutrients in the diet. So specifically we can also consider things about local foods or wild foods and how they might contribute to adequate diets and understand diet quality and diet diversity. We can also help us to understand identify areas of concern in the diet or identify population groups of concern in the diet. So we can we can investigate things like healthy and less healthy dietary patterns and also investigate the link between diet and health. But we can also use data to evaluate or consider the need for food fortification programs to monitor and inform national food policies, guidelines and health education programs and track the monitor progress toward towards the SDGs. Additionally, we can look at dietary shifts. So what we see for example is a shift towards ultra process foods in some countries or some regions. We can assess the sustainability implications of food choice. And finally, we can use dietary data to monitor and inform food safety policies and agricultural policies. So with these reasons in mind, I'd like to share with you all some information on the global individual food consumption data tool or gift. And this was developed by FAO in collaboration with the World Health Organization, together with other international partners and is funded by the Bill and Melinda Gates Foundation. This tool serves as a platform to make global individual quantitative food consumption data publicly accessible available from all countries around the world, collected through both large national surveys and also small scale surveys. The platform provides easy to understand infographics on food based indicators for nutrition and food safety and also allows users to download micro data. So on the side in front of you is just a summary of the metadata that we have available on this platform, which is up to almost 300 surveys now worldwide. And we have micro data available for 24 surveys, as you can see from this map. And in order to be able to share infographics and micro data, we first need to harmonize this data. So with this in mind, I wanted to just highlight some of the considerations that we think about when we consider food identification. And I'm sure many of many of you here today will be familiar with these points and I know that we've already heard some of these mentioned in the previous present presentations. So we may have a very long food list for a country and I think the one that Judy mentioned for the Caribbean is a perfect example of this. A very long food list that may need to be constantly updated. And this is because diet is so varied and we can observe such a huge variation in the food, the foods consumed between countries but also even within countries. We now see a greater variety of foods and more diverse foods, more international cuisine as well. More processed foods are being consumed and new foods are appearing on the market every single day. The recording of homemade recipes remains very complex. And as I'm sure you're aware, we need to be recording all ingredients of a quantifying those as well and even ingredients like oils and water recording the cooking method and the cooking time. And obviously then vitamin, mineral and water losses need to be accounted for. But not only that foods may have multiple local or international names or be even incorrectly named by locals. And it's for these reasons that there's really a need for harmonization of the food description and the food code of consumed food items. There's also a need to maintain comparability while not losing the precious detail information about what has been consumed. So this slide is just to show a few examples that perhaps you've come across already where we may face challenges with regard to food naming, food descriptions and food coding. So there's, we see often in dietary surveys examples of where maybe the same food is recorded, but it has a different name. Or maybe a different record different food is consumed, but it has the same name, or maybe a food that has an incorrect name, but it's well known by local. So here are a couple of examples. So the cookies and biscuits often used interchangeably aubergine and eggplant. Also, palm apple, fruit or ice apple, then all the confusion around chips fries potato chips fried potatoes crisps potato crisps, which is such a challenge. And then other other things that come up often in surveys and I've just picked this one example here from a ongoing work in the South Asia Biobank survey, where we were really struggling to correctly identify in the dietary assessment, whether or not the meat consumed was mutton or goat. And this is specifically because of challenges in local naming of foods. And so now moving on to food next to food next to is a comprehensive system allowing the classification and description of foods. It's developed and managed by the European Food Safety Authority or FSA and food next to is an approach to establish a common language across databases worldwide. So the same food can therefore be referred to in different ways. And the example here is for a food car key, as it's known in Denmark. Excuse my pronunciation, if that's incorrect, but also the person as it's known in the US. So food next to we're able to assign a common code as show. And we would be able to do this for the for the other examples that we saw on the previous slide for example the aubergine and the eggplant. But it's important to note that the food next to code does not replace or substitute any previously assigned food code or food description is actually there is an addition. And this is purely for the purposes of harmonization. More information can be found on this, this website of FSA. So food next to allows us to precisely describe the food through the use of facets, which are descriptors that can be added to the food next to base code. So for example, if the car key from the previous slide was consumed with peel, it is possible to add a facet to describe it. And there are 32 facets, not all of them are used but many of them are very useful. We have facets for example describing different processing that have food is undergone before consumption, or others describing the fat content of a food or identify fortified foods and inform on the fortifying agent. So all can be added to the food code and carry the information to describe the food more specifically here on this slide as an example. So one useful point to mention about food x2 is that you can adapt the level of the coding according to the detail that you have about food. So for example, if we only have the food descriptor as milk, we can actually code that as food x2 code is shown on the slide there. However, if we know more about that milk, for example, that it was a cow milk from a cow, there's a further code that can be used. And if we have more descriptors we can add facets to describe the additional information included in the food description. So as you can see here with the most detailed milk description we can assign a code for milk from a cow with the fat percentage specified. And food x2 is therefore a very comprehensive and flexible system for food coding, especially since we know that there can be a huge range of detail that can be recorded in dietary surveys. And some details may have been omitted, for example, if the participant didn't know the precise type of milk that was consumed. It's also interesting to mention that the codes included in the food x2 are organised in a hierarchical way. There is more than one way of organising and grouping the food x2 codes and this is really dependent on the needs of the user and the type of data being coded. So you can see on the right hand side here several different hierarchies and the hierarchy known as exposure is the one that is used to code food composition and food consumption data. This is a screenshot of the food x2 catalogue browser. Which is the tool developed by EFSA to support the coding with food x2. The browser is openly available and can be downloaded from the EFSA website along with further documentation on its use. Coding with food x2 may require some training, which can be provided by EFSA to European countries and by FAO FAO to other data managers interested in coding food composition and food consumption data. Next, I wanted to show you the process that is undertaken with the data owners when they want to share their survey data. The first step for us is to identify the existing data, contact the data owner and validate the basic criteria, for example the sample size and the method of dietary assessment use. Then there is a very important process of post harmonization because the data set needs to be mapped to the food x2 categorization system. Sometimes the food x2 coding is carried out by the data owners themselves after training by FAO. And then this coding is then checked by the FAO team. In other cases the FAO team carry out the food x2 coding. For both cases there is a lot of interaction needed with the data owners. And then at the very end of this process the data can be shared on the FAO WHO gift platform. So moving on to the harmonization of food composition tables with food x2. The increased harmonization of food composition data with food x2 can actually enhance the consistency and reliability of nutrient intake assessments. So there's several food composition tables that have been already coded using food x2 and they have been disseminated through the in foods website. So those include the food composition table for Western Africa, the food composition database for Phytate, the global supplement database. And additionally we have been carrying out some food x2 coding on other food composition tables and they will be disseminated soon including the U pulses, the U fish and the Indian food composition tables. The necessary elements to identify, describe and classify a food item from a food composition table are shown on the slide. And they include the food name, the food description, the food code and the food group and food x2 allows for this. This slide demonstrates that how if on the left different coding systems are used, the food x2 matching needs to be done manually. And the risk of error is much greater. However, on the right when one coding system is used, a large part of the matching can be auto-automized and therefore the risk of error is diminished. So FAO is contributing to the scaling up of the use of food x2 to the global level. You have one minute left. Okay, thank you. We have an ongoing collaboration with EFSA and WHO, and we have contributed to the adding of food items not consumed in Europe. So for example, plants and animal species from other regions. We've contributed also to the addition of wild food items such as insects and edible flowers, and we have a regular exchange with the team at EFSA on newly identified food items or coding issues. We have facilitated an upgrade of food x2 through the analysis of existing FAO, WHO give data sets and there are also regular updates of food x2 catalog by EFSA. FAO undertake regular capacity development activities and to date over 200 data managers have been trained by FAO in data categorization using food x2. We now also have 24 dietary data sets harmonized and shared on the GIF platform. So just the take home message from myself is that the nutrition assessment team uses, advocates and actively supports the harmonization of individual quantitative food consumption and food composition data using food x2. And we are working towards harmonizing those data and sharing those data as widely as possible. Thank you very much. Over to you, Ruth. Yes. Thank you, Richard for bringing us near the use of data of dietary data and how the food x2 is working and what it is. This is really, it is a system as Richard said and can. It's really very important if more data sets from the food composition and as well as the food consumption side, if they would have the same coding system up to level two or three or four, even. They can have an automatic match if they are the same or very similar. So, so this will be a huge advantage for the future to to have and and also the reported foods in some kids and navies they will also be coded with the help of Bridget's group. So that food matching in the future will be much more easier. So, having said so and again you know a big big thank you to each of the presenters. And we are now open for the questions and we have received several questions. The first one is, like always, or like very often if this is recorded yes it is recorded, and we have on the inputs website in your category which is the webinars and they are in, in some time, it will be published the recording of this, of this, of this session. Share presentations I think you have to contact the person if they are willing to share the presentations. Then, I would like to go for Judith and for banana. Is it necessary to match according to water content for high quality food match. Do you want to try to answer that one. If you want you can go. Yeah. Oh, okay. I think one of the errors that I have seen in older surveys where we don't have the wonderful tools that food X will become for us is where people have matched a dry food to the nutrient data for a prepared food or vice versa. And we do get very large errors in the food matching and in the estimation of nutrient intake so there's some very, it's important to think about is the food that I'm matching to a dry food or as eaten food. It's always possible to tell in a food list and certainly in this project that I was involved in. Mostly we could tell but not always. And depending what area you're working in you may have a lot of knowledge to guide you or you may have very little knowledge to guide you. So one example of where we use moisture content to validate recipe work, we had a separate project of calculating the nutrient content of local recipes in the from the region. And we estimated from that a moisture content for the food based on the weight change of the food during cooking. And when we looked at those moisture contents it was a very good guide to whether the recipe recording had been done accurately. Because if you had a food that was ready to eat like a stew, and the moisture content was very low maybe 25 grams per 100 grams. You knew something was wrong in the calculation because no one would eat the food like that. So I think I personally have used the moisture content as a guide to the accuracy of recipes and to whether I am matching a dried food to a prepared food. Fernanda you may have some suggestions for other ways that it's important. Yeah I just want to compliment I think that checking the components is very useful when you are looking at the food composition table and the food description is not complete. So for example if I have a, if you find a food described just as being and it's not saying if it's fresh or dried you can just check the moisture content. Or if you are checking a prepared food and it's not mentioned for example a vegetable if it was added a fat to this preparation you can go to the fat content and also check. I think that looking into the food composition table sometimes because these details are missing in the food description we need to compliment with the checking of the nutrient content of the certain food. So I think that this is, and how Judy mentioned are the ways that we usually apply the nutrient values to check against the food description. I think if your survey is reporting on sodium intakes. Looking at the sodium value will be a very good guide as to whether the data refers to a pre salted food, or one that has no salt added because an unsalted food will have a very low sodium content in most cases. But often the name of the food and the description in particularly older databases may not say whether it's salted or not. So again, for Judy and Fernanda. How to take into account the difference in crops from different countries when you use all this data for the Caribbean. Yeah, that's a wonderful question and it's something that's very difficult to do in a real life project. So thinking about the Caribbean data that we prepared. So we may have had, for example, data for Taro, and that's grown in the American continent, and we only had data that for Taro that was grown in India or in New Zealand. Very different areas of the world, very different soils, so the nutrient values may be different, the cultivar may be different, the color of the crop may be different. All that in that case, all we could do is by indicating the country from where we got the data. And any assumptions we had to make and quality about the match is that users of the data later can look at all that information and make a decision about how good quality, how reliable an estimate based on that food might be. I mean, ideally, you, if you were looking at crops, someone might actually be in the country and go to the market and look at what is available, what color is it, how ripe is it when people buy it, how does that match to the data that I have available. In reality, when you're doing a project such as the one we just did, this is not possible, so you have to make some assumptions. Exactly. So, not as good data is always better than no data. So I think this is the bottom line of everything that we do and this is where we have the quality, the very good, the good, the medium and the low, so it's better to have low quality data than no. The next question is also to you both on how come imputing missing values. It's a whole, it's a whole lecture. So can you just give two or three things, but I think you have given already a lot but you know just one or two, because if not it will take too long time. Just a few very simple ones. There's many foods for which you can be confident that a nutrient won't be present and you can make an assumption zero. One way you might do it is say I only have data for beef, but I want data for goat. I will, they're both ruminant mammals, I will use the data for beef. So it's, as Ruth said, it's a very complicated process and that's why we couldn't complete this part of the project because we just had, it's too takes too long to do, but you're looking for similar foods, total absence of a nutrient. Maybe a food similar food that's been cooked in the same way. I think, yeah, really this could be a separate talk. So, the next question, food lists keep it small. If it comes from dietary survey data, do we need to confine them before matching? I think that this is a really difficult question to answer. I mean, in, because it depends on your survey and how the project has been operated. So for example, in the Caribbean survey, the food list had already been prepared and put into the software. So it was very difficult to reduce the size of it, even though as the food matches we would have liked to have less data to work with. It's very hard to change it once it's already been put in place. And I was always unsure whether this was just a very long list of potential foods or these were things that people had actually reported. So it's, it's quite difficult and also a software system is usually done before the survey happens. So you have to make assumptions about what people might eat. And you don't want to lose a lot of details. So then you end up with a very long list. Some things that you can do to shorten it to think about, is there the same food present many times, just with the name slightly different. And we found that there were many cases where exactly the same food, maybe just the word order was changed because maybe different people had compiled the list. So there's some simple ways you can reduce it just by having a bit more time to refine it. But in all real world projects, you don't have a lot of time. And you have to make some decisions and you have to proceed without knowing everything you need to know, because otherwise you will never finish your project. And as Ruth said, if you have no data, it's no help to anyone. It's better to have some data than to have none at all. Yeah. Yes. Thank you so much. Let's go for the next one. Food case. Is it free of charge and how to get it? And how to take the nutrient values into account doing food matching with food case. If there are several matches in food case allows it to calculate means. So these are the three questions for Karl. Yes, okay. I hope I can remember all three of them. Let me try. Yes, let's open the first one. Yeah, so I did not show the slides. So there is a food case.org website and where we have now started also a little bit more promoting because until now we had kind of keep it a little bit secret this food case, but we're promoting it now. So food case.org is a website where you can get more information. And the users of food case, it's open source. Yeah, so we built a software in an academic area. So for sure it's all open source and full case is basically also for free. We just have in Europe some rules and which basically we need to discuss, but that's just for for the European partners. And the second question is the food composition data during the matching. So we did not import from global diet, which basically have some nutritional values. We did not import them because we did first at the beginning we didn't know where this food composition data came from. And later we found out but then we did not use it for the matching. So the food composition data get only on the food composition data site. So we just had the food names and the descriptors in for the matching. So we did not use them in this way. And the last one is this is a very interesting question. So if you have multiple matches and what we're doing. So we were also considering but in Iran, in such a matching run we have only one matching. So we said, okay, let's work with one because otherwise, yeah, we have to make this decision what to do. We are kind of outsourced this problem into the food composition table. So we have one this a bridge and show the wonderful example with the cookies. And so for instance, we also said if somebody's mentioning a cookie in a consumption interview and but doesn't know exactly any details. So we kind of aggregated 400 cookies into an average cookie. And then we just say okay then in such a case if somebody you have no idea about the details. We just say okay match it to the average cookie so we're kind of extend the food composition database have it with average cookies average vegetables cooked vegetables and raw vegetables and so on and so on. So we just to have there a more appropriate match so we wanted to increase the quality going this way. But of course, yeah, it's it's kind of a challenge so that's maybe something for the future that let's say we're also allow multiple matches. And then, but then the question is how you decide maybe do you wait them the same way or do you take one more than the other so market share suddenly comes into play right if you can try to find out how it's a more market share of certain products then you can also adjust the waiting so the more the more market share the more waiting it gets. But this data we did not use so this because it's another complex dimension and some also costly data if you want market share data you have to buy it and they're quite expensive. So we did not include them this way. Okay, so do I understand correctly that in the next version or in one of the next versions, you will add that you can do your mean of several food, food entries from a food composition table and probably waited for the moment now. For the moment, yes, we just can do it in food composition in this part, we can do it so that you have it there but I'm not not in the matching run. Yeah. Okay. Thank you. So then for Bridget, so how many data sets in the future do you expect to have in gift. Yes, many thanks for the question so by October next year. So, by October next year, we should have at least 58. That's our target and then we can hope to continue after that. Great achievement. Then there are another some more questions so can I use the food composition of a raw food for a cookbook. Sorry, no, go ahead. Ideally, no. If you have time. It's better to find the data for the cooked food cooked by the method that with your food relates to. If you don't have it, ideally you would calculate it using nutrient retention factors and moisture gain or loss. So the decision that we did use raw ackee fruit data for cooked ackee. We had to make that decision because it was very late in the project and we were nearly out of time and we just did not have the capacity to do any more calculations. So, yes, please, if you if you can please calculate it for cooked and use raw as a poor quality outcome if you have no other alternative. Yeah, because the cook the composition of raw food and the cook food is very different. So, if ever you would try to avoid that, especially for highly consumed foods, you would never do that. So if it is a few food which is consumed once the error is on the population base is is is insignificant. So it can be done and this is what they have done. So, yeah, so in general, never do that only on an exceptional base. If you are running out of time or or a budget or whatever, and only for foods which are really not consumed a lot. And in the food composition database, you would never do that. And then we have another question for green leaf of vegetables so how do I do it with indigenous one. And can I calculate the if this green leaf of vegetables in a dried form, if food composition is available on the fresh form, say amaranth leave, but no data on tried moringa leaf is there a way to compute the values on the tried for. So it's not on food matching it's on how to calculate. So, yes, there are some methods on how to calculate if you know, especially the water content and then you can use the retention factors to to decrease in the in the green in the green ones. So for the food matching so what would you do with the food matching if you have green leaf of vegetables. How do you do it with some of the indigenous ones. I think that when we don't have any data for the original green leaf vegetables what we usually do is to take an average value of other green leaf vegetables and I think that is important to consider especially if they are dark green or light green, because this will change a lot to divide many composition. And it's not like the, the best approach but it's better than leave in blank so if we don't have any any other data available, usually when we are making a food composition table, we try to look for data in the literature. I don't know if we have published data but if we don't have I would for matching data from food consumption survey but the composition table I would suggest an average value of other green leaf vegetables. Thank you. So the last question so would be it would be interesting in a food grouping from food surveys is also assessed during matching food composition danger. I'm not sure I understood this question but probably somebody else did and is able to answer. I don't know if I understood the question very well with but I do see a couple of other questions about. I could answer if that's helpful. Okay. So there was a question about the food matching the facets for food and a query over the understanding of that is that done only through the food name. So apologies if that wasn't clear from the presentation. So we do use the description primarily. But if there's any doubts on the description, and we would, and food composition data is available we would also look at that. So I hope that clarifies that point sorry that wasn't clear. Okay, and Richard, do you have some example on how you deal with chips and fries. Yes, I saw you also saw that question thanks Ruth. And so this is a good question and I think it depends on the stage at which you're at, because this type of things should really be dealt with during the data collection. It's stage. So if you already know some of these types of concerns and for example this example on the mutton and the goat. This was something that was then addressed to interviewers in the field on who are doing the data collection about the importance of understanding exactly what it was that the participant consumed. When it gets back to us it's, and the data is already collected. I think we have to look specifically at things like the country that the data is coming from, and, you know, kind of make a judgment on that, basically. So expert judgment. Yes, yeah, if it is not precisely described in the food consumption. Okay, then some more questions to call. Do you use the food tag in in food case and then I may add and input partners. Another question to call is, are you planning to develop an R package for food matching, or no of a similar initiative. Yeah, thank you. Good questions. Yes, basically food case we have now for New Zealand so I also saw in the audience and we have it with the input tag name so basically we have a version which is only using the input tag names. And we have a European version, let's say where we have this euro fear tags which are very similar to the inputs but work slightly slightly different right so basically this is there. Other taggings is just we're using food next to language and own classification that the country normally has on their own right so this is the one. And with our package is interesting of course are is very popular. And for the moment, no, yeah, because we're having that the food case system as national databases. And so we're had we needed to be wanted in Switzerland for sure and also in other countries that the function in there. And otherwise we need first to export the data and do this and in our so it's kind of not not planned to do it there yeah so, but I also don't know from similar initiatives basically where the question. Let me maybe add something, there are several questions and if you have a raw or a cooked food. What we also had experience with someone when we had this tool, and we had a number of mentions so we have seen the tab watch was mentioned about three and a half thousand times, and also other food so if you see a food was very often mentioned. And kind of followed the approach okay try to be there very good in the matching right so make a high quality matching because that the food is very often mentioned. And if a food was mentioned only once or twice let's say in a big survey, you can say, okay, so if I don't have data, I can either investigate or just say okay and I make an approximate matching let's say right just with a low confidence let's say so. And this was kind of a very nice approach I saw the good strategy that say that the highly mentioned foods, maybe you should also check how much it contributes to the whole diet so tab watchers of course very important. And but you have so maybe some spices on this. And this is mentioned a lot spices is in comparison to the rest on the plate it's not a big part right so you can also say there and maybe also should consider how much it contributes but the number of mentioned was a very good indicator for the simple way to achieve high quality analysis there because then you look first of the most important ones and match there very well. And the other ones, you can maybe with no confidence matching is kind of fine to go. Okay, good. So, I think we are all in. Okay, the same message, which is very good and we reinforce each other with our understanding and the presentation. So we have one question left before I will then close and this is for calm. Can this food case, can you estimate food flavors. Yes, I've seen this question food flavors so I understand the taste right yogurt if it's vanilla or strawberry right. Yes, it is the chemical. Food flavor. Yeah. Okay, no yeah. Interesting topic. But not really. Yeah, so I take this home. I was kind of surprised to see for that. It's a good question. So, not yet, because it's not on food composition tables and it is a different database which I think doesn't really exist. So, in theory, yes, if you have the database with the components of food flavoring yes you could do it and then you have to import your, your database and then you can do it. Yeah. But with the actual databases that we have no it would not be possible. So, I would like to thank everybody the presenters really for excellent presentations and the other participants so at the end, we have had 126 participants out of the 280 registered. So, and, and thank you so much for your, your questions and your participation, and I hope that you found it as interesting as we did. And then we have more seminars, more webinars to come. We have a webinar on data evaluation on the 12th of April on indigenous food on the 21st and on fruits and vegetables composition on the 15th. So, of April. So, probably you find this thing seems as well interesting and looking forward to see you again. And again, thanks so much to every presenter and to all of you who have been present. Thank you, Ruth, for organizing this webinar. Thank you for this opportunity. Yes, everybody would like to add something. I would like to add to Ruth, I think that soon you'll be moving to a new stage of your career and I'd like to thank you for the wonderful effort you put into food composition over very many years. And I think everyone who's involved in food composition really appreciates it. Thank you so much. Anybody else would like to make a remark before we close. Let's also say thank you to you for all the organization and all your life work. Yeah, likewise. Thank you, Ruth. And we're looking forward to the next webinars. You're not going anywhere yet. Okay, so thank you very much. And this is the end of the seminar and thanks again to everybody and looking forward to see you in other webinars. Bye bye. Thank you. Bye bye.