چكيده به لاتين
In the growth of social networks with presence of huge product and services choices for users, the recommendation systems have a prominent role in finding interested items for users.
In this thesis we focus on movie recommendation and propose two topic models to finding similarity of movies by modeling movie plots and genres. The models find topics that have been well structured in term of movie contents. Meanwhile we use CTR model that can be factorize user-item matrix to learn user and item latent features, respectively. The CTR can use topics, which achieved from topic models, to learn item latent features appropriately.
So our proposed model find topic proportions in movies and use it to feed CTR for recommendation task. We use movie plot corpus to learn topics and movielense-10M100k rating dataset that is 98.2% spars, large scale and highly imbalance dataset, to evaluate CTR model. Finally we show that modeling of movie genres, in addition of movie plots, can improve CTR in tem of recall and precision in out-of-matrix prediction.
Keywords: Recommendation Systems, Topic models, Bayesian Network
