Research On Leveraging Overlapping Community Impact For Personalized Recommendation

Information overload has become one of the prominent problems in the era of big data and personalized recommendation has become a hot research area that has been proven to alleviate information overload and enhance user experience.The existing recommendation method focuses on the interactive data between users and items.However,the scale of users and items is tremendous and the user-item matrix is extremely spare.So the existing recommendation method utility the interactive data as the sole data source may suffer cold start and data sparsity problem.To solve these issues,researchers consider introducing other auxiliary information,such as social relations between users,commodity labels,etc.Few people consider that the overlapping community can accurately reflect users ' preference.In this paper,we propose a recommendation method considering overlapping community effects based on the fact that overlapping communities can effectively reflect users' preference.This method combines interactive data between user and item with overlapping community to infer individual preference in topic model which is sharp tool for semantic mining.The method regards the users as documents,items and communities as two types of words,and users' preference as topics and transforms the decision-making process into the generation process of words.Based on the principle of topic model,the method can model user preferences and infer the model parameters by Gibbs sampling.Finally,the utility value of items is calculated according to the parameters,and the items with higher utility will be recommended to the target user.Empirical studies based on the real word datasets demonstrate that our methods outperform several state-of-the-art algorithms that utilize communities' information in item recommendation.