Research On Forum Content Recommendation Approach Based On User Interest

With the popularity of the Internet, the Internet has become the main way for people to obtain information, bringing great convenience to people's lives. However, the explosive growth of information resources also brought the problem of information overload, personalized information service system in a certain extent, solve this problem by analyzing the user's interests and browsing history, recommended to the user information they are interested, greatly reducing the user time to find useful information. Currently, recommender systems have been widely applied to e-commerce, community forums and other applications. However, the current personalized recommendation system is not very mature, thus there is a recommended low accuracy, timeliness and poor scalability issues.In order to alleviate the situation above, this paper proposes a new social forum content recommendation method based on user interest. Firstly, we define the keyword category and build models for users and posts based on keyword category. Secondly, by analyzing user behavior in the social forum, we define the calculation to measure user's implicit score to visited posts. To tackle the drifting of user interest, the forgetting curve for user interest is introduced. By combining the score function and the forgetting curve together, we can get the final calculation of current user interest to viewed posts. Thirdly, the paper predicts users' scores to unviewed posts using an improved post similarity calculation based on content similarity and user interest similarity. Finally, an improved calculation of cluster centers is proposed and the neighbor selection is performed on the post clusters and user clusters, then the final predicted score is calculated for the target user. The results of recommendation based on user clusters and post clusters are combined for the final recommendation list.To evaluate the effectiveness of the proposed method, the paper conducts a series of experiments on the BBS data from Northeastern University. The proposed model construction can reduce the dimensionality of the vector and contains some semantic information, thus it can save the cost and increase the accuracy of model building. Integrated calculation of time function and user ratings can improve the accuracy of user's current interest to viewed posts. The proposed adaptive selection of initial cluster centers can be automatically generated and number of the initial cluster center, thereby improving the clustering results. The improved post similarity calculation can also make similar posts results more accuracy. The combined results of post based clustering and user based clustering neighbor selection recommendation results can make recommendation results both relative and novel. The comparative experiments show that the proposed social content recommendation results based on user interest is better than traditional recommendation method.