Study On Context-aware Recommendation Based On Trust Mechanism And Its Applications

In the information age, the huge information brings much convenience to our dailylife. However, the problem of information overload also comes following theconvenience, which makes people cannot find the information that they wanted. Then,the recommendation system emerges as the requirement. Nowadays, CollaborativeFiltering recommendation System, which is one of the most successful and widelyapplied, is proposed to help user filtering the information Recommendation Systems.Since the traditional Collaborative Filtering recommendation System does not concernmultiple standards in recommendation, it sometimes performs poorly in projects withmultiple standards. Besides, fake ratings, losing people’s interests in different situationalso make Collaborative Filtering recommendation System cannot meet the people’sneed. Then, how to improve the quality of recommendation systems becomes asignificant event. This paper makes the further research for the RecommendationSystem based on the analysis of the existing recommendation systems.Firstly, since traditional recommendation system does not concern the multiplestandards in project, this paper proposes a multiple standard recommendation algorithmbased on Widrow-Hoff. The algorithm improves the quality of recommendation byusing the user preference function which is derived by Widrow-Hoff, and the high levelGoodness of fit of LSP in Widrow-Hoff.Secondly, this paper proposes a method to generate the confidences between the usersfor fake and vicious ratings existing widely in the recommendation system. Thealgorithm improves the quality of recommendation by concerning the confidence andsimilarity in recommendation.Thirdly, this paper finds that the user’s preferences are changing and therecommendation results should be different in different situations. So, our algorithmparts the user rating time into several time periods and makes recommendation on eachtime period. Then, the algorithm makes the final recommendation by integrating all timeperiod recommendation with their weights.At last, the results of many experiments indicate our algorithm is better than otherrecommendation algorithms. Future work is written in the last of paper.