Research Of Personalized News Recommendation System Based On Time-ordered Behaviors And Tag Relation

The huge amount of news in the Internet,which is updated in real time,meets the needs of different users.According to the latest statistics from the China Internet Network Information Center,83 percent of Internet users are accustomed to reading news on the Internet.As more and more users access the Internet via mobile devices and wireless networks,how to help Internet users find the news they are interested in from massive news has become increasingly necessary and urgent.As one of the most important methods to solve the overload of Internet news,personalized news recommendation technology has been highly valued by the industry and academia,and a number of related algorithms and systems have been proposed and developed,which makes the research on personalized news recommendation algorithm become very active in the field of data mining and machine learning.Based on user behaviors and content of news,personalized news recommendation algorithms use various data mining techniques to analyze user interest and recommend news to the user.Personalized news recommendation algorithms can not only help users find interesting news,but also make user be satisfied with the website.The traditional personalized news recommendation algorithms rarely consider the time sequence characteristic of user browsing behaviors and ignore the value of the tags of news which contain rich information.This paper does some research on personalized news recommendation algorithms by considering the time sequence characteristic and tags.The contribution of our work is as follows:(1)The user browsing behavior is a kind of time series data,which has the time sequence characteristic.However,traditional personalized news recommendation algorithms rarely take the time sequence characteristic of user behaviors into account.So that the performance of traditional personalized news recommendation algorithm is poor when they are used to recommend the next news to user.In order to solve this problem,this paper takes the time sequence characteristic of user behaviors and the context of user into account and proposes a personalized news recommendation algorithm named time-ordered collaborative filtering.To avoid the defect of traditional user similarity measure,this paper proposes a user similarity measure called time-dependent similarity measure to compute the similarity between a long-term user and a short-term user in a more reasonable way.(2)Although the tags of news are related to the content of the news,traditional personalized news recommendation algorithms always use a keywords vector or topic distribution to represent the content of news,which ignores the function and value of the tags of news.In order to distinguish the importance of tags,based on the amount of information of tag and the degree of tag in the probabilistic tag graph,we propose a method to calculate the weight of different tags.Based on the co-occurrence of tags,we measure the relativity between two tags in the way of conditional probability.Based on the probabilistic tag graph,we use a tags vector to represent the content of news and user's reading preference and propose a personalized news recommendation algorithm to recommend news articles which are closely related to user's reading preference.(3)Based on our personalized news recommendation algorithm,we design and implement a news recommender system.The news recommender system not only can recommend user news articles they are interested in,but also recommend user news articles that are closely related to user's reading preference,which can increase the diversity of recommendation.In addition,to demonstrate the efficiency of our personalized news recommendation algorithm,extensive experiments are conducted along with detailed performance analysis.