Research And Application On Collaborative Filtering Algorithm

With the development of Internet,the information grows incredibly.People have to consider how to filter right information from vast infor-mation.Recommendation technology is one of the effective methods to solve information overload.In the past decade,a larger number of re-searches on recommendation technology emerged,and many kinds of recommendation algorithms are proposed.Collaborative filtering is the most successful and widely used method,with important values on re-search.Main work taken in this thesis is as following:(1)This thesis studies the recommender system,summarizes the re-search background and status of the recommender system.Meanwhile,this thesis introduces and compares the popular three type of recommen-dation algorithms,especially on collaborative filtering and then gives a detailed introduction on the implementation on key algorithms,similarity measure and evaluation metrics in collaborative filtering.(2)This thesis proposes two advanced MF models,namely CW-MF,NICW-MF respectively.CW-MF considers the importance of the item categories,and NICW-MF considers both the impact of item categories and user neighbors.Experiments on MovieLens datasets show that two models perform better than comparative methods which demonstrate a user considers the item categories and the advice of his neighbors when he selects items.(3)This thesis proposes a probabilistic model named NUIT(Neigh-bour Users Impact Topic model)for group recommendations which sim-ulates the group item selection generative process.NUIT considers not only the group topic,user's personal interest,but also the member's neighbours with high similarity to make group recommendations.NUIT considers above three factors.Experiments are conducted on two datasets,and the results show that our method performs well.(4)This thesis designs and implements a parallel algorithm based on Spark to compute the user similarity.MovieLens dataset is used to evalu-ate the algorithm performance.The results show that distributed compu-ting can improve the computing speed obviously on large data.