Research On Parallel Collaborative Filtering Recommendation Algorithm Based On User Potential Relationship

With the rapid development of information technology,more and more information is flooding people's lives."Information overload" has become a hot topic of the world.In order to solve this problem,personalized recommendation systems have emerged.The video recommendation in the IPTV platform and network video platform is one of the important application fields of the personalized recommendation system.In order to improve the user's viscosity and improve the service quality,quickly and accurately recommending videos of interest to users from massive video libraries has become an urgent problem for current video service providers.Collaborative filtering recommendation algorithm is a classic personalized recommendation algorithm.Its main advantage is that it does not need to consider the content information of the resource,has the ability to discover new information,and the design idea is easy to implement.However,this algorithm also has problems such as sparsity,timeliness,and scalability.This topic is based on the video recommendation field,analyzes the problems in the field,and conducts relevant research.The main tasks and innovations of this topic are as follows:1)Through the preprocessing of user historical data,extracting user behavior features to form user behavior data;Aiming at the problem that implicit rating model in IPTV video recommendation cannot fully measure user preference,a multi-feature mixed implicit scoring model based on video type,bias factor and viewing ratio is proposed.Experimental results show that the accuracy of the proposed model is improved by 9% compared with the ratio-based scoring model,which can fully reflect the user's preferences.2)An improved collaborative filtering recommendation algorithm based on the user's potential relationship is proposed.The algorithm uses the random walk method to calculate the indirect similarity of users to mitigate the influence of sparsity on the accuracy of the recommendation,and uses timing impact to resolve user preferences changes.The problem of timeliness and the combination of indirect similarity and time-series influences provide a comprehensive measure of the degree of similarity of users by merging similarities.Experimental results show that compared with the user-based collaborative filtering recommendation algorithm,the proposed algorithm reduces the recommendation error by 11%.3)The collaborative filtering recommendation algorithm based on the user's potential relationship proposed in this paper is implemented on the Spark parallelization platform,and the impact on the algorithm is mitigated by expanding the computing nodes in the cluster.Experimental results show that the parallel implementation of the algorithm has good scalability.