Page 115 - AC/E Digital Culture Annual Report 2014

AC/E digital culture ANNUAL REPORT 2014artificial neurons, can predict whether a book will sell well or not. Its mathematical model is able, they state, to make estimates using multiple variables, such as price, points of sale, publisher, etc., together with the economic situation at the time the book is launched, the author’s name, literary fashions at the time, etc. This system also grows with the accumulation of data together with its ability to learn and adapt to new data. Its creators say that the margin of error is barely 18%. I do not know whether any publisher has so far used these services.language in reference both to actions and to descriptions, for example, of parts of the body. On the other hand, it seems that, according to this algorithm, bestsellers base their language more on verbs that describe processes of thought (remember, recognise) rather than actions and emotions. The report gives a number of grammatical formulae, verbs and nouns that appear to be most used in works of this type. In fact, novelists who write in a more journalistic style have greater literary success. It is not the only study. Another study5 claims to have found an “emotional algorithm” able to analyse millions of words with emotional density in literary series—the test is based on the works of Shakespeare and the brothers Grimm—but is also applicable to any tool for textual communication, including the Web and social media.The analysis of language already has commercial applications, as in the case of Luminoso, which understands and analyses different languages semantically in real time. What it basically does is to analyse language to determine whether a particular product or service has really brought satisfaction so as to be able to recommend other services or products with a similar level of real satisfaction. By understanding the data it has about users, it can create a recommendation system that goes beyond just the sales made. A cultural entity, for example, can find out what people really say about its work or what it is that users really want. The technology can even distinguish from the semantic context whether it is a book or a film that is being discussed in those cases where a film has been made of the book with the same title, and whether people liked it or not. Its algorithm, which is fed daily, adds all kinds of terminology, slang, metaphors and whatever may be the figures of speech and linguistic registers used on the Web to gradually refine understanding of what is being said and people’s feelings and opinions. What underlies the words is the real answer from consumers or users, which is shown in a graphic interface that looks like a cloud of complex tags in several semantic layers. There are those who assert that, on the basis of these algorithms, it will be possible one day for computers to write novels,Nonetheless,the creators ofanotheralgorithm statethat they havefound the key topredictingwhether a bookwill be successful based on “statistical stylometry”, that is, a statistical analysis of the literary styles of various genres that identify those characteristic stylistic elements which are most common in best sellers in comparison with those that fail to achieve this status. The research4 is based on 44,500 books in the public domain published by the Gutenberg project. The researchers counted as bestsellers those which had been a critical success and had been downloaded a large number of times from the Gutenberg Project website. In addition they included others such as A Tale of Two Cities, by Dickens, The Old Man and the Sea, by Hemingway, The Lost Symbol, by Dan Brown, and the latest Pulitzer prize‐winners together with some “super sellers” at Amazon. The algorithm analyses parts of sentences and the use of grammatical rules together with another type of semantic analysis.Some striking data emerge from the report, such as that bestsellers make more frequent use of conjunctions and prepositions in comparison with books that are less successful. Also, the research found a higher percentage of nouns and adjectives in bestsellers than that found in less successful books, which depend more on verbs and adverbs to describe the action of the plot, and more negativeAC/EAlgorithms to predict publishing successes or to analyse millions of words in literary works with emotional densityWHERE WE ARE HEADING: DIGITAL TRENDS IN THE WORLD OF CULTURETHEME 9: THE NEW AFFECTIVE TECHNOLOGIES COME TO THE CULTURAL SECTOR CURRENT PAGE...115