Abstract:Recommendation is an important and also challenging problem in online social networks. It needs to consider not only users’ personalized interests, but also social relations between users. Indeed, in practice, users are often inclined to accept recommendations from friends or opinion leaders (users with high reputations). In this paper, we present a novel recommendation framework, social recommendation using global rating reputation and local rating similarity, which combine user reputation and social similarity based on ratings. User reputation can be obtained by iteratively calculating the correlation of historical ratings of user and intrinsic qualities of items. We view the user reputation as the user’s global influence and the similarity based on rating of social relation as the user’s local influence, introduce it in the basic social recommender model. Thus users with high reputation have a strong influence on the others, and on the other hand, the effect of a user with low reputation has been weakened. The recommendation accuracy of proposed framework can be improved by effectively removing nature noise because of less rigorous user ratings and strengthening the effect of user influence with high reputation. We also improve the similarity based on ratings by avoiding the high similarity with the less common ratings between friends. We evaluate our approach on three datasets including Movielens, Epinions and Douban. Empirical results demonstrate that proposed framework achieves significant improvements on recommendation accuracy. User reputation and local similarity which are both based on ratings have a lot of helpful in improvement of prediction accuracy. The reputation also can help to improve the recommendation precision with the small training sets.

Improving Rocchio Algorithm For Updating User Profile In Recommender Systems

Improved Metric-Based Recommender by Historical Interactions.

A Novel Framework To Process The Quantity And Quality Of User Behavior Data In Recommender Systems

A User-based and Item-based Collaborative Filtering Recommendation Algorithm

RØROS: Building a Responsive Online Recommender System Via Meta-Gradients Updating

SOCIAL INTEREST FOR USER SELECTING ITEMS IN RECOMMENDER SYSTEMS

User Interest Dynamics on Personalized Recommendation

Incorporating Community Detection and Clustering Techniques into Collaborative Filtering Model

Collaborative filtering recommendation algorithm based on user correlation and evolutionary clustering

User Profiling Based on Keyword Clusters for Improved Recommendations

Research on Collaborative Filtering Recommendation Algorithm for Improving User Similarity Calculation

Prediction on recommender system based on bi-clustering and moth flame optimization

User profile correlation-based similarity (UPCSim) algorithm in movie recommendation system

A New User Similarity Model to Improve the Accuracy of Collaborative Filtering

An Improved Collaborative Filtering Algorithm based on Dimension Reduction and Improved Clustering

SoRS: Social Recommendation Using Global Rating Reputation and Local Rating Similarity

DRprofiling: Deep Reinforcement User Profiling for Recommendations in Heterogenous Information Networks

The effect of low-level image features on pseudo relevance feedback

Online Recommender System Based on Social Network Regularization.

Collaborative Filtering Recommendation Algorithm Based on TF-IDF and User Characteristics

Social Recommendation Algorithms with User Feedback Information.