Friend Recommendation Using Probabilistic Matrix Co-Factorization

Abstract

Nowadays, the number of users on social network sites or e-commercial sites becomes much larger and users’ information on those sites are normally heterogeneous, so friend recommendation becomes more and more an important issue. Many researches have been proposed such as the graph-based approach or content-based approach or hybrid ones, however the model-based method is few, especially the one can utilize a variety of user information. Advancing previous work, in this thesis we present a novel model-based algorithm whichcan incorporate both the friendship graph and the user rating matrix to learn sensible user descriptors for making friend recommendations. Then, for the experiments we use the benchmark Filmtipset dataset and prove that our algorithm outperforms the base-line methods.