Recommender system based on NLP: a support tool for the publishing sector
In recent years, the Spanish publishing industry has been getting closer and closer to digital transformation, however there are still many challenges to overcome, especially in recommendations to the end customer (readers). With the advancement of Machine learning, a branch of Artificial Intelligence, it is possible that many of these challenges can be met.
However, it is important to note that the books will have to be recommended by a literary genre individually, since they present completely different characteristics from each other. For the analysis of Non-fiction books, this work will be based on historical sales and the author's influence on social networks. That is why, at the architecture level, it will be defined as a boosting model, which will reduce errors in predictive data analysis. And for the analysis of Fiction books, it will be based on the emotions that the text transmits. That is why, at the architecture level, it will be determined by Transformers networks, whose responsibility will be to manipulate sequential data, mainly in the field of NLP. Finally, and at the design level, the outputs of these subsystems will serve as input data for the recommendation engine.
In this thesis, a final recommendation system based on Natural Language Processing (NLP) is proposed, whose main contributions are: (1) the integration of multidisciplinary professions such as psychology, literature and artificial intelligence; (2) a design of an intelligent system that recommends based on the behavior of the reader and not only on their purchases; and (3) boost literary quality while still attracting people who don't usually read.