Improving Recommendations by Using a Heterogeneous Network and User's Reviews
Vítor Rodrigues Tonon, C.C. Oliveira, Daniele C. Oliveira, Alneu de Andrade Lopes, Roberta Akemi Sinoara, Ricardo Marcondes Marcacini, Solange Oliveira Rezende · 2019
A recommendation system is an information filtering technology that seeks to predict and recommend items to its users through an analysis of their history of interactions with the system. An important point to consider is the analysis of user's reviews, which allows the aggregation of valuable information in the recommendation process. However, commom data representations for recommender systems usually are not enough to capture all the relationships between the entities of the problem. Therefore, in this work, we propose a heterogeneous network that aggregates information of users, items and reviews in a single representation for the recommendation task. Using the proposed network, recommendations were generated by a network regularization method. The experiments showed that the proposed method is very promising.