Latent class estimation models and application to Movie Recommendation

Abstract

Models that estimate latent classes for movie recommendation based on PLSA and decision trees are proposed. Proposed model can explain the reason why such recommendation results are given. Using proposed model for so-called cold start problems in recommendation, we can handle the users who don't have enough records. Instead of conventional PLSA for recommendation, we use decision tree models consist of some questions. So, instead of using user's records, we can recommend suitable movies using user's answers as input of decision trees. In an experiment of questionnaire survey, improvement of the satisfaction of the proposal is 45% in comparison with the previous method by showing the recommendation reason. Another experiment is implemented where the users who have less than 9 movie viewing are recommended more appropriate movies after answering 5 questionnaires.