Research On The Recommendation Model Based On Theory Of Reasoned Action And The Problem Of Cold-Start Items

Owing to the information overload,there are challenges that helping modern consumers choose among so many options and showing the contents to the appropriate people.Recommender systems have emerged in response to this problem.To date,recommender systems have been applied successfully to a wide variety of fields,such as movies,TV shows,books,music,news and so on.Nowadays,recommender systems attract the attention of industry and academia because of the immense business value.However,most of them recommend items by predicting preferences.In fact,the behaviors play a more important role than preferences.Inspired by theory of reasoned action,we proposed a recommendation model TRARM focusing on user behavior.Our method make recommendations by behavioral intentions.Theory of reasoned action shows that user behavior depends on their behavioral intention.And the behavioral intention is related to the person's attitude and subjective norms.Specifically,the attitude presents the user's valuation of the consequences of performing the behavior and the subjective norm is considered as the influences of relevant individuals.In TRARM,the preferences traditionally utilized to generate the recommendations can be seen as attitudes in the TRA,while the subjective norms include the other users' attitudes or reviews of the items.TRARM combines both of them to provide recommendations for users based on the behavioral intentions.Additionally,we propose a method NRLCS to mitigate the cold-start problem on new items by learning network representation on items-attributes graph.NRLCS learns the latent representations of item vertices which is used to obtain user vectors.Then the recommendations can be made based on the similarities between item vectors and user vectors.Compared with other cold-start solving methods,there is no need to collect additional information in our approach.The sample survey results and experiments on MovieLens show that TRARM is more reasonable and effective.Furthermore,the experimental results demonstrate that our method NRLCS is effective and superior.