Estudio de técnicas de aprendizaje automático en un recomendador basado en crı́ticas

Abstract

[en] This project is focused on improving critique-based recommender systems based on critiques, which allow the user to indicate the features of a desired product through feedback in order to provide his preferences and receive products that satisfy them. Specifically, the application of unsupervised learning systems to the recommendation systems based on critiques is studied in order to improve the efficiency of the critique-based recommender system. More specifically, in this project, the technique of clustering is joined to the History-Guided Recommender, which is to date one of the best critique-based recommender. From this union emerges three new recommendation algorithms: HGRCUMsesions, HGRMaxCluster and HGRSameCLuster. The efficiency of the new recommendation algorithms is analyzed and it is observed that the efficiency of these new algorithms improve previous recommendation algorithms. In addition, there is a large number of clustering algorithms so it’s necessary to study if the efficiency of recommender systems based on critiques that use clustering depends on the clustering algorithm used. In particular, we will deduce in this project that, using the CUM recommendation algorithm, similar results will be obtained using different algorithms of clustering.