Design of a recommendation system for books

Artificial intelligence is increasingly playing a crucial role in helping users find products that are well suited to their interests, tastes and wishes. Nowadays users deal with a huge amount of data when they want to purchase or borrow products. The aim of recommendation system is to help the user in his decision making process, driving his focus on product that fits his needs. Recommendation systems are models capable of suggesting personalized items to a generic user by means of filtering techniques. This tools have acquired great relevance and have been widely used in recent years as they represent a significant improvement, especially in relation to the Big Data issue. We can find their application in disparate areas, they play a fundamental role in various services: from the entertainment such as movies, videos, music, to the e-learning web sites such as newspapers. Furthermore, they are vital to improve the online e-commerce on different products. One of the main aspects of a good recommendation system is the reliability of the recommendation results, this increases the trust level of the user in the system. The thesis work consists in the application of literature recommendation techniques orienting them to the world of books. Specifically, the application scenario is to recommend interesting or popular information suggested by the community of social network aNobii. The word of mouth on social networks among readers with similar tastes is crucial to extrapolate possible suggestions for the online book trade. In addition, an analysis on library loan data was carried out thanks to the collaboration with the libraries of the city of Turin. It was possible to analyze and merge this data-set with the aNobii social network data. The application of the recommendation techniques on this merge result data-set allowed to study the users behavior and therefore to suggest them a good book to read suitable to their preferences. Aggregating big data is not easy challenge, due to the huge volume of data is a feat that must be overcome efficiently through essential steps. A fundamental part of the work concerns data preprocessing and cleaning. The investigative analysis on the data is of paramount importance as it permits to transform them into a more understandable, useful and efficient format for their use. This important task allows to have significantly better performance in the subsequent steps. During the work, collaborative filtering approaches of recommendation models were applied. The results obtained showed a significant improvement between the models implemented respect the baseline model. In particular the betterment between the best performing model and the baseline model was approximately 38%. In parallel, it was possible to perform a distinct further analysis to evaluate the similarity between different items by exploiting the content of the books metadata and using similarity metrics. Finally, it was interesting to classify and identify the opinions of users on the books reviewed with the help of sentiment analysis techniques. Social networks contain meaningful information, but at the same time it is difficult to extrapolate them. The sentiment analysis aims to capture the opinions relating to a single item and exploits them to improve even better the suggesting strategy.