Collaborative Filtering Algorithm Research Based On User Context Information

As is known to all, with the rapid development of Internet technology, we haveentered the era of information overload from the lack of information. However, it ismore and more difficult to obtain the goods with which they are satisfied.Recommender system is considered one of the most important ways to solve theproblem of information overload by providing personalized service to usersautomatically. A complete Recommender system consists of several components,while the most important part is recommender algorithm. Collaborative Filtering isconsidered to be the earliest and the most widely used recommender algorithm. Thealgorithm provides recommendations of unvisited commodity to the target user fromitem recommendations recommended by the user who has similar interest. Thetraditional Collaborative Filtering algorithm does have a high-quality of accuracy.However, there still exist many worthy improvements. For example, the traditionalCollaborative Filtering algorithm could not effectively solve the problem of user’sinterest drifting. On the other hand, The traditional Collaborative Filtering algorithmspecially focus on the user’s similarity but ignoring the user’s own characteristicsThe user’s context information plays an important role in RecommenderSystem. This paper specially selects the time information and the trust information asthe user context information to analyze the impact on Collaborative Filteringalgorithm based on the user context information.Using time information includes the user preference model and the user interestmodel. User preference refers to their own characteristics, which means different users are not with the same score standard when they rate their favorite items, andthe standard of the same user is different during different periods. The change of userinterest with time is called drafting of user interest. Generally, the user interest isvarious and changes with the time, so there are long-term interest and short-terminterest, which have different effects respectively. Therefore, this paper designs anovel way to reflect the changes of user’s time information, which been able toadjust the short-term interest weights and the long-term weights. Experiments showthat Collaborative Filtering algorithm with user context information is better than thetraditional one on the recommendation accuracy.The paper also proves that the trust information of the user plays a veryimportant role. In our real life, when we need some information to recommend someitem, we would like to turn to the users who have similar interest with us. On theother hand, whether the users could be trusted also needs to be measured. In thispaper, we argue that the trust also can be calculated. On this basis, we create a trustmodel that can calculate the trust value of the users. Then, we combine trust modelto traditional Collaborative Filtering algorithms. Experiments show that trust methodcan lead to an improvement on predicting recommendation accuracy.Finally, we design a hybrid recommendation algorithm with theabove-mentioned two kinds of user context information, which uses a weightedapproach to achieve that. Experiments show that using a variety of contextinformation can get better results than using a single one.