Algorithm Research For Collaborative Filtering Base On User Rating Preference And Local Item Space

With the rapid development of Internet technology,the human has entered the information society and the Internet economy era.That makes the number of users and items growing rapidly.It is difficult for users to find favorite items.Therefore personalized recommendation system come into being a role.It can provide recommendation services for user's personalized requirement.Collaborative filtering is the most widely used and successful recommendation algorithm in recommendation system.But it faces with some problems,such as data sparsity,cold start,algorithm scalability and still needs to improve the accuracy in some situations.In our research,we focus on the related problems and the latest achievements of the collaborative filtering recommendation algorithms.Based on users' rating preference factor and users' local similarity factor,which ignored by related researches,we put forward the improved algorithms to address accuracy problems.In this paper,the research content mainly includes the following two aspects:Firstly,based on the fact that different users rating items with different rating preferences,we propose a collaborative filtering algorithm based on users' rating preference(Rating Preference Distance,RPD).The algorithm aims at mitigate the influence which causes by user's rating preference in selecting neighbor users.The algorithm distinguish ratings from two aspects,which are “positive” and “negative”,and construct a preference factor formula from the two aspects.Then we use the factor to improve user similarity measure in order to improve the accuracy for selecting neighbor users.We also reconstruct calculation formula for prediction rating from positive and negative in order to improve the accuracy of prediction rating.The comparison experiment shows that,in the final of score calculation,the RPD algorithm can greatly improve the accuracy of prediction ratings and lay a good foundation for selection of items to be recommended.Secondly,based on the fact that there is local similarity among users,we propose a collaborative filtering algorithm based on local item space(Local Item Space Distance,LISD).The algorithm aims at improving the accuracy of recommendation by calculating user similarity and prediction rating in a local item space.The algorithm distinguishes the similarity among users more particularly from the angle of items.At first,it chooses a local item space of a target item to be recommended before selecting neighbor users.And then,it uses improved similarity measurement to select neighbor users in that local item space,and compute prediction rating in the same local item space.Finally,we globally compare calculated prediction rating in each local item space to generate a recommendation list of items.The comparison experiment shows that LISD algorithm can raise the precision of prediction rating and accuracy of recommendation results.