 So, application in musicology. So, the idea here is, okay, we have another graph. We can explain it, not using the embeddings or all this stuff, more much in relating to MIT. Here, just the graph. I'll see if we can get something useful for musicology. It's probably a graph. So, musicology embraces study of history, theory, and practical music from many ways. And musicology is part of the humanities. So, musicologists have to read a lot. Musicologists have to read a lot of documents, be in the library. So, what is the computer? It's able to read all for a musicologist. It gives a summary, analytics, whatever. Solve the Western dialect. So, this is the idea. Naturally, this understanding can help musicologists. So, we build, like, three different applications. It's not limited to this, just what we have done. Entity relevance, analytics, and information solidization using the graphs. So, we have the graph, and we can get them to be relevant with analytics or visual analytics. So, entity relevance. So, if we did a graph of entities, as the one I explained before, so we apply entity link into every document, and we create this graph, connecting all the entities. So, for example, remember the biography of Wilco. We read the graph, Wilco to alternative broad, and the platform, and so on. So, you can see, you can imagine that this is a, these are hyper links. And if you click here, you go to alternative level with the library. Go to the application to library. So, if you click here, you go to the restriction of alternative broad. If you click here, you go to the data library. So, it's like we're hyper links. So, we can use algorithms for, that are used for search engines to rank these using the network of hyper links like Google does with the page track. So, we build first the graph of entities, and then apply page track. That's the idea. And get the relevance of them, we have a support, not in the graph, and then we classify to rank them. Okay, this is the most relevant thing. So, in a paper, it means, in 2015, we build a knowledge base of flammable music. We gather information from different sources. But then we have like, about 1,000 biographies of flammable artists. And we apply it to these biographies. And we read the graph of entities from these. Set those biographies and apply to the page track. And it's unbelievable to see what are the most relevant singers, guitar players, and dancers in flammable. And we ask a flammable expert to give us a list of the top ten artists in each of these categories, in his opinion. He told us, oh, it's very difficult to have a unique list of artists. But okay, I will elaborate a list of ten artists that might be in a list of top ten. So, he did this list, and we ran the algorithm also, and we compared them. And in the top five, we get very high precision page rank. So, this is the list that the page rank give us as the most relevant singers, guitarists, and biographies dancers. And it has a lot of sense, really, that the mythical singers, guitarists, papaluchia, and eternal dancers also. So, it means that what the idea of applying page rank over these works. You can get these elements. You can get artists or other entities, we did also in the dictionary, and get elements of schools of music and conservatories in the world. You can get the elements of any entity. So, another thing we can do analytics. So, we can extract whatever we want information extraction on a guitar library. For example, we can get artists, edidimensions, sediments, relations, sediments, whatever, and we can do different kind of analytics. So, I just give an illustration of some of the things that you can do. So, we did a research with Rob Nicianard, that is one of the largest record works in musicology. And we get about 16,000 biographies and we did the process of extract information. We did the editing, but we also extract some specific information like the role of the artist, the birth and death plays, and they we also did relation structures and biographies. We did some analytics. So, here are the birth dates and it's the number of the birth dates, the number of artists every year. So, there are some tendencies that perhaps can illustrate the musicology and gives some ideas to the musicologist. We also run the number of the role of the different artists in the digital age, most of them are composers, teachers, conductors, we also try to analyze the migratory tendencies of artists. So, we have the number of artists that was born were born in a country and they were born there in that country and also in cities. What was interesting in single file was that in Paris, it seems to be a very nice city to go in the last days of your life. So, this was extracted from text so the digital age really has brought it and if you do some analysis, you can see inside that. So, the idea in the field is that the musicologist can even ask a question and the system understand everything and get the answers. Okay. We did another study of analytics study. It's a chronic study of effective albums. So, the same dataset that I told before we use for classification from Amazon Previews. So, we have this dataset that has like, 66,000 albums and we have more than 250,000 biographies. No biographies, no reviews, so, these reviews come with a start rating even by the user who did the review and we have for every album we have again a map tag, we have metadata, what is the artist, what is the label so we have this information with music rates and we get also the gear of publication of the albums. So, at the end we have some information and what we did is we applied sentiment analysis to the customer religion. We applied the sentiment that is in every review and then did some average in terms of gear of publication and gear of where the review was written. So, this is aspect-based sentiment analysis, that is a kind of sentiment analysis where in sentiment analysis you typically do like justification. You have a text and you give a label in positive, negative or negative but this aspect-based sentiment analysis what does is to okay, we have a document but perhaps there are different things inside the document with different sentiment so, this is what it does. So, the idea here, in a sentence we can have entities, we can have aspects that are like features that are described in the text that he can produce some vocals in this example, we can have opinion words, great surely, so great little piece so, this is an opinion word that is referring to this aspect and all this is referring to this entity with contract and it's a song so, this is the opinion. So, what we did is to do that on other reviews and get all the aspects and the opinion words and give us score. So, at the end what we have here, we have a review with a set of aspects like guitar, vocals, sound we have a score between minus one for every one of them like saying how much the review is positive inside this aspect or layer. So, using that well, this aspect based on different analysis is based on so, based on a sentiment lexicon so, this is score is based on a sentiment lexicon so, a sentiment lexicon is a dictionary where you have a lot of words for each word associated so, you have great for great you have one you have good and you have 0.7 you have bad and you have minus 0.5 or whatever so, you have other words and every word has a score so, using this dictionary of scores the idea is we did it by first search nouns in the review, like noun phrases and also, I gave this which adjectives are noun phrases and what is the score of this adjective there is a negation between what is the distance from the adjective to the aspect so, using all these stuff with some rules we made these scores for a review for the aspects in a review so, then what we did is like we arranged all these scores for a review and then we get like a score of the review between minus 1 and 1 and then we arranged all the reviews in two different ways we take arranged them by the review obligation here in where this review was written by the customer when, when was written and by the publication here of the add-on so, when was the add-on published this review so, this review is about when was the add-on and when was the add-on published so, classifying averaging the reviews in terms of this so, this gives us an idea from a customer perspective and this from a musical perspective so, here is the study by review obligation here so, when the review was written so, we have the data set of Amazon is from 2014 so, here we have the average score of the of the same different analysis for, here for example, all the reviews published in 2010 this is the average score of all the reviews so, we have this and this is the average rating that the user is because we have every review has a rating between 0 and 5 stars given by the user that so, this is the average of all the scores given by the user for all the reviews given this year, for example so, more or less, we don't see any correlation between the sentiment and the rating so, perhaps we think, ok, perhaps the sentiment is related with another thing with the economy, perhaps so, this is the GDP of the United States so, this is the economy, not in the United States for example, this is an hypothesis or, what happened in the United States because, we see that in 2008 we have a high peak, what happened in 2008 well, there was the election of Obama perhaps this was like, people have a lot of hope and use more positive adjectives in general in the United States, perhaps this is really good, I know we should see what happened here by the way, this is just an hypothesis and we don't provide any support but, we did also as we have also the general labels for the albums they are spread by general so, all the albums all the country albums all the jazz album, all the music now, we see that most all of the genres has been in 2008 but country doesn't have it that's also when the economy crisis yeah, yeah, that's what we mean which I said before, not this economy crisis 2008? 2009 so, and if we have an H5 aspect, so as we have in every review, as whole for different aspects like song, beat, whatever almost all the aspects has this pipe in 2008 so the reviews has slightly more positive adjectives talking about them that means so, well, for the studies necessary to evaluate any of these suggestions correlations as an incapsacea but they are necessary for physical needs to study these so, we also did a study from the point of view of the year of publication of the album so, here for example, we have the average sentiment of all the reviews from albums published in 1990, even the review have been published in 2014, for example this review was about an album that was published in 1998, so what are the reviews and here, the rating so, here we see really a strong correlation between the sentiment and the ratings from this point of view we compute personal relation and then 0.75, it's pretty high so, there is a clear correlation between the sentiment and the ratings it means that perhaps we may use the sentiment to predict the rating and also that the sentiment has kind of sense, because it is correlated with the rating so, we also did this study of the the average of the sentiment by geograph, because we have information on geograph, so this is the reviews of pop and the reviews of pre-albums, and this is the sentiment scores, and we see that there is a peak in the 60s for pop, in the 70s and the 80s for rating, so there are more positive attitudes in the reviews of album published in this year and the same as this, so perhaps the interest is influencing this positiveness, or the and on the golden age of frame is here, so we can start the evolution of geograph using that ok, finally you can visualize it if you have a geograph you can visualize it and that can be very useful for the rating documents, so you may have a digital library, you have a version extraction, and then you are able to navigate to the library using the geograph, you have this geograph, and we need some proletary work with that