 So good day. My name is Arindam Basu. I am an associate professor of health sciences and I am going to talk about my presentation on using computational text analysis to filter fast pass title and abstracts in systematic reviews. So let's get started. Now, one of the things that we know about systematic reviews is this that they're quite time consuming processes. Although systematic reviews are excellent for generation of evidence and therefore this is a lot of interest. A lot of people do them. One of the most important point where we spend a lot of time is incorrect selection of primary studies. And selection of primary studies are essentially on the basis of titles and abstracts. But before you do that, you need to have a protocol and then you need to have some sort of a way in which you can search the studies in literature databases. And then you select the studies on the basis of which you want to build your systematic reviews and meta analysis. But after you obtain the, your total number of primary studies, it can often go up to hundreds. And sometimes I have seen in my previous experiences about several thousand studies needs to be analyzed on the basis of their titles and abstracts. This process is time consuming and it is prone to errors. So one of the ways in which people do this is this that if many of us, if a team is conducting the systematic reviews, then we have open our protocol. We select keywords from our inclusion and exclusion criteria. We read each title and abstract carefully. And on that basis, we select whether the, this is this, this particular title and abstract fits our purpose or we have to reject it. And on the basis of this, we eventually create what is known as a Prisma chart. And this goes into there. So can we use computational text analysis or quantitative text analysis where we can use a computer and algorithm to go through the process and make it faster. In this example, we will be using a particular type of quantitative text analysis software called Quant EDA, which was written by Ken Benoit. It is a free package, which is, which you can install in our software and you can work. So we used a study where we wanted to see what is the effectiveness of mindfulness meditation for anxiety relief. We ran a search on the, on, on, on PubMed and this, for this example, we are only going to use PubMed and we found that there were 95 studies. There were all English language human studies and we could, we could, we got that study in, in a, in a form of CSV file. So then what we did was this, that we loaded Quant EDA, read text packages and tidyverse libraries onto R. And then we used this particular version that read text function in order to read the base file, which contained the URL of it. And we assigned the text field, which was title, the text field will be the one which we processed. The next step was that we conducted, we converted that PubMed object into a corpus and the corpus contains all the essential information along with the text level, along with the document level variables in it. Then we converted the entire corpus into a set of tokens, which means that we split up the, the individual sentences, which were the title words, titles, essentially into individual words. And then what we did was this, that we used several options so that we removed common words, which were like stop words, English language words such as a and the etc. Then we conducted or then we created our own dictionary, which contained all the terms that were included that, that we wanted to include in our inclusion criteria. So that we could put together our inclusion and exclusion words that the software then go and search in the titles. And then what we did was that that we created a document term matrix or document feature matrix in particular in case of quant idea these terms are referred to as features. And then we used a, we use the dictionary that we created earlier with the document feature matrix. And then what we did was that we looked up the terms and classified them with the belonged to the, to the inclusion criteria that we set up earlier. So this to two tables, one was a feature stable that you can see here. And so these are the terms mindfulness randomized meditation anxiety. These are the statistics that you get to see that you can see that from here we see that there were 46 studies. And that were randomized control trials, the ones that we wanted to see 55 out of 95 studies contain the term mindfulness 22 studies contain mindfulness based. So there's a slew of around 77 studies that contain the terms mindfulness, but not all of them were randomized control trials. And we see that the outcome term anxiety was found only in 25 out of 95 studies. So you can see that very quickly it helps us to narrow down on the terms that the studies that you want to see. So in terms of the outputs, you can see that using combining these various terms, we are able to generate the outputs that we would like to see and select the studies that would be useful for us. This is only a header that I'm going to show you. So then you can export the outputs to filter studies based on the basis of titles, you can see how they can be rapidly narrowed down. You can use that on the abstracts, not just on titles. And computational text analysis can be used for even full text articles, such as PDFs. I must say that this is very much a work in progress. If you want to take a look at the GitHub repository, you can go here and you can work with us. Thank you.