Research Of Cross-Domain Recommendation System For Multi-Source Data

With the development of intelligent networks and devices, information on the Internet is much richer and diverse, people can obtain a variety of information and services on the network, at the same time, people gradually realize that to find things on the Internet in line with their needs has become increasingly difficult.In order to solve this problem, people used search technology in the early days, but because the resultsfromsearch engine were not enough "personalized", can not meet the different needs of the users' personal preference, so personalized recommendation technology began to be proposed and used in various fields, recommended system also try to use various algorithms mining the user's individual preferences.Combined with the current situation of widespread multi-source data, to alleviate the data sparsity and user cold-start issues, raised multiple models. For sparsity of single domain data, put forward a variety of multi-source data cross-domain recommendation model, namely the integration of user similarity matrix, user predictions, the global behavior of the user to predict the score. For users cold start problem of single domain in multi-source data, the recommended model based on collaborative filtering and cross-border mining association rules in the traditional cross-domain collaborative filtering based on rules related to the second filter.After test the proposed model on the Douban dataset, we can be found that compared to the traditional single-domain recommendation algorithm, information fusion of multi-source data model improved cross-domain recommendation on the recommended effect, can effectively alleviate the sparsity of single-domain data. When new users to recommend, cross-domain recommendation algorithm association rules compared to traditional popularity and cross-domain recommendation collaborative filtering, improved the prediction accuracy, effectively easing the user recommendation system cold start questions, so it can be used to enhance the effect of the current recommendation recommendation system.