The Research And Implementation Of Collaborative Filtering Recommendation Based On Context Clustering

Along with the progress of the rapid development of wireless network and mobile device,network resources are explosive growing,leading to the problem of information overload.Recommended systems have become one of the most effective ways to solve this serious problem.Recommended algorithm is the core of personalized recommended systems.In these algorithms,Collaborative filtering recommendation is one of the most successful recommendation technology at present.It completes this task by analyzing the users' preferences which are highly similarity to the target user,and the project which neighbors like will be recommended to the target user.Though collaborative filtering,recommendation can mining users'potential interests.On the other hand,there are also some problems in collaborative filtering recommendations such as data sparsity,scalability and cold start and so on.Besides,user score matrix does not fully reflect the interests of users.Their interests also have the link with age,education level,occupation and many factors.People who have similar attributes are also prone to similar interests.To solve these problems,this paper proposed the collaborative filtering algorithms based on the context.This research completes the following work:1.After deeply studying the main recommended algorithms,analyzing the problem of context loss.To address this problem,this subject puts forward the user context information and its formalization combined with Situation Semantics and the factors impacting the interests of users.2.To solve the sparsity problem of collaborative filtering recommendation,this thesis puts forward the classification method of user context through further study for user context.After deeply analyzing the dissimilarity degree calculation of different types of variables,building the dissimilarity degree matrix.What's more,after studying fuzzy c-means algorithm,this subject puts forward a improved FCM by using a convergence factor to solve the convergence problem,which will cluster users in different groups accurately.Through this process,M×C×N turns into M×N,which can solve data sparsity.And also this proposal can reduce the complexity of the algorithm.3.To solve the problem of ignoring differences in scale in a single score in the traditional collaborative filtering,this thesis puts forward an improved similarity algorithm,which brings in a balance factor.This factor will relieve this problem,and help user find their neighbors accurately.Then in this subject,we combines this improved similarity calculation with clustering based on context.This algorithm will provide users with more suitable recommendation.4.Finally,using the method proposed in this paper to experiment on MovieLens dataset repeatedly,and comparing it with collaborative filtering recommendation based on equivalent context dissimilarity matrix clustering by MAE,precision and recall,which prove the method have better performance.