Research On Bursty Topic Recommendation Algorithm Based On Topic Model

With the popularity of the Internet,social networking platforms such as Facebook,Twitter,and Weibo have rapidly gained popularity due to their convenience and speed,and them have gradually become an important source of information besides traditional media.Many users use social platforms to express their feelings and share interesting content and topics with friends.At the same time,many bursty topics on social networks are often released on the social network platform,and spread quickly through the user's friend forwarding mechanism,causing widespread social resonance,which in turn has a huge social impact.The problem people face is not only how to track the bursty events and public opinion in the massive Internet information,but also to find the real-time topic information that they are really interested in.Therefore,a recommendation algorithm that can detect the unexpected events and push them to interested users could make a better user experience.In view of the above requirements,and some thesis based on the previous research,we uses Sina Weibo as the research platform,improves the traditional topic detection technology and proposes a microblog recommendation method based on the topic model.The prime work of this paper has the following three parts:Firstly,according to the bursty characteristics of burst events in Sina Weibo,the burst word feature is used to detect the bursty topic in the microblog,and the topic probability distribution of the burst microblog is obtained by using the LDA topic model.Among them,the relative word frequency,word frequency growth rate and burst weight are used to screen the burst words,and the microblogs with the burst words are marked as burst microblogs,and the topic of the microblog is simulated at the same time.The experiment proves that the actual collected microblog text data not only improves the topic detection efficiency,but also reduces the influence of noise microblogging.Secondly,according to the different user behaviors in Weibo,the user's different interest level of Weibo theme is proposed.An LDA expansion model is proposed to integrate user behavior.The model will be based on different user operation behaviors of all Weibo texts of the same user.Gathering different user documents,and using Gibbs sampling method to estimate parameters,the feature word vector in the user document is converted into the subject probability distribution of the user in different behaviors,thereby reflecting the user's interest preference on different topics.Finally,the information aging theory in information metrology is introduced in the collaborative filtering recommendation algorithm.The recommendation weight is adjusted by integrating the user's interest degree and information time efficiency parameter.Compared with the traditional recommendation algorithm,the improved recommendation method is more time-sensitive.High,can recommend microblogs with bursty nature to interested users.