Research On Recommendation Algorithm About Topic User In Social Networks

Along with the rapid development of Internet and the emergence of social network, information has been growing explosively, traditional information searching model is already not suitable for modern situation. Recommendation system arises at this moment. Personalized recommendation service in Social network based on users’ interested is currently a hot research area, such interesting researches includes user interest mining, early warning and public opinion on hot news, social circle discovering. One of these directions is personalized recommendation system.Although has certain achievements in the field of personalized recommendation, recommended quality of traditional recommendation algorithm in information flow of the social network such as weibo is not high because of the complexity of social network in different background. And also due to the different recommendation purpose, there are different ways to optimize. Through the analysis of the existing recommendation algorithm and the characteristics of topic model, we put forward the topic user recommendation model.The main work includes the following several aspects.First,due to the prevalent problem of incomplete data and low efficiency in social network, and also because of the different requirements for data acquisition between different recommendation algorithm, in this article, we present a distributed crawling framework based on Sina Microblogging, API according to our research needs. Under the various constraints from the calling of API interface, we develop a reasonable strategy interface calls in order to get more and more comprehensive data. According to the system diagram and data crawling flow diagram, we can see that the system is use of the parallel way to crawl data and the extension of system is very strong, it also can solve the problems about that the authorization code automatically validate when it expired and API call automatically sleep when its call completely over.Second,the model use LDA topic model as a powerful tool for the latent semantic mining on text; its biggest difference from the traditional text model is that its text model is established on subject level, we can get documents-topic distribution through the training of LDA model. We get each user’s weibo-subject distribution through the trained LDA model, and translate it into user-subject distribution to depict the distribution of the user’s interests to realize the mining of user’s interest. This paper also processes a standardization on micro blog. Build multiple dictionaries such as expression dictionary and Internet user dictionary through weibo data structure. The text pretreatment and segmentation effect is remarkable.Third,studies show that theme users have the widespread characteristics of outstanding theme and big influence, In order to make accurate extraction to the theme user in Topic User Recommended model, The topic user extraction model in this article are extracted through the user-topic distribution characteristic and the user’s interested rate, so as to build a decision tree by this two main characteristics to differentiate ordinary users and topic, ordinary users as recommended target, subject users as the main body to be recommended.Fourth,in order to improve the accuracy and diversity of theme user recommendation,topic user recommendation algorithm combined the similarity of users based on content with the correlation analysis between subject, as well as topic of authority users topic to recommend, through certain strategies this algorithm fuses three candidate sets and get the final TOPN topic user hybrid recommended model.Experiments of weibo data and test results show the feasibility and effectiveness of the mixed recommendation model. According to the evaluation method of recommendation algorithm, compare the results of hybrid model with the results of random recommended, the effect is improved obviously, the diversity and accuracy of performance is pretty good.