Modeling Of User Interest Based On Topic Model And Its Application In News Recommendation

Information overload has become a serious problem for Internet user. Mining content of interest or information of value from massive news for users has become more and more difficult. Moreover, with the boom of mobile internet and the access of large numbers of intelligent terminal, the requirement of precise newsfeed is increasing. The utilization of personalized news recommendation system can effectively help to handle the problem of information overload and improve users ’ reading experience by providing personalized reading list. Online news reading has become one of the most popular web application currently.The key of personalized news recommendation is personalized modeling of user profile, which extremely depends on precise and integrated characterization of user reading interest. As both news and user reading interest have topic characteristics, in this paper, with the introduction of topic model, we perform the topic modeling for the user interest in order to achieve personalized and intelligent news recommendation.In the aspect of precise modeling for user reading interest, this paper proposes a parallel hybrid recommendation method with a fusion of content analysis and user collaboration in order to solve the problem of personalized news recommendation. Taking account into the characteristics of multi-topic in news, this paper utilizes topic model to perform the topic modeling for the user reading interest, and then perform the precise modeling for the user by secondary topic modeling for news reading user, and lastly combine the analysis of the timelines of news to improve the quality of recommendation on the basis of the above. Experimental results show that, the hybrid recommendation in this paper is better on the precision indicator of recommendation than two baseline methods which are as a comparison, and have excel ent performance in diversity indicator.In the aspect of integrated modeling for user reading interest, taking into account the fact that user reading interest is distributed in different scenarios of application domains, for example, user’s interest point may be distributed in the scenarios of application domains such as Weibo, QQ, Wechat and Douban, based on that, this paper proposes MTM(Modified Topic Match), a cross-domain transferring model of topic interest. This model, b y fusing the user’s topic interest in Weibo into user’s reading interest in the news reading domain, achieves the characterization of multi- source integration of user interest,and thus improve t he integrity of the user modeling. Experimental results show that, a cross-domain transferring model of topic interest in this paper, not only avoid the problem of user cold-start to some extent, but also obtain certain results in mining unknown user reading interest, which greatly reduces the costs of getting the user reading interest.