Research Of Collaborative Filtering Recommendation Algorithm Based On Clustering Expert Selection

After the wave of web2.0 development, the information of today's world tend to be more diversified, complex and specialized. To find information the user wants in the mass of information, in the 1990 s, there have been personalized recommendation system. Personalized recommendation take the user's behavior and data to mining the user's needs and interests.Personalized Recommendation into content-based recommendation, collaborative filtering recommendation and knowledge-based recommendation. Collaborative filtering algorithm is the method which the most widely used. Against the insufficient accuracy problem of Collaborative filtering recommendation, main work is focused on:1) Propose a new method of field division. Divide the field of commodities, allowing the system to quickly find the target commodity-related areas to understand users. How to divide the field, it was suggested that artificial classification of goods, to find neighbors in the area of commodities in commodity-related objectives to recommend to the user. But artificial classification is not reliable and could not find the potential relationship between commodities. Therefore, we propose to divide commodities by clustering methods. This can help the system to identify potential relationships between commodities, take the potentially associated ratings as references to recommend to the users, especially in the situation that goods can not be artificial.2) The introduction of a trust factor in the similarity of this formula. Traditional similarity measure by the similarity between the user's score. The problem with this is that the user can not determine the reliability of scoring,the scores may not be reliable. To solve the problem of unreliable user ratings, this article will trust as a factor in the similarity measure. The advantage of this is that we can weed out those scores who are not reliable,thus algorithm recommended more accurate and satisfactory.Based on the above, we propose collaborative filtering recommendation algorithm based on clustering experts selected. The improved algorithm first clustering to divide the project area of commodities, next find experts in the related field,thus we can ensure the experts have a better understanding of the related field. In determining the expert users, the trust is introduced into the similarity formula, considering the similarity and trust of users. Therefore, the algorithm selected expert users with similar interests and high reliability for the target users.In the experiment, we use the prediction accuracy and success rates to measure the effect of the algorithm. In order to verify the stability of the algorithm, we measure the accuracy of this paper algorithm and user-based collaborative filtering recommendation algorithm under different trust environmental. The results showed that: the accuracy of collaborative filtering algorithm based on clustering expert selection is better, it give the customers satisfaction results. Especially in the hostile environment prevailing rates, the effect of this algorithm is also more stable. The accuracy and stability of this paper algorithm are better than the user-based collaborative filtering algorithm, therefore this paper algorithm has a higher value.