Research On Weibo Recommendation Algorithm Based On Topic Model And Fusion Similarity

In today's diversified information,online social media has developed rapidly.Making more friends and learning more interesting topics are the main needs of social media today.Sina Weibo,as a new form of online media,has gradually become the main platform for people to share daily life and obtain real-time news.There is a lot of available information in Weibo data.At this stage,Weibo users are limited to receiving the information they care about,while other hot topics and topics that may be of interest are not actively pushed.Therefore,how to recommend such information has become an important direction of Weibo research.The content of micro-blog data is short,the semantic information is scarce,and it has a high degree of sparsity.Traditional topic models cannot fully extract useful information.And because the traditional Weibo recommendation does not consider factors such as time and similarity,the recommendation accuracy is relatively low.In response to these problems,this article mainly improves Weibo recommendation from the following two aspects:(1)When directly using the LDA model to build a Weibo user model,there is a short Weibo text length and a lack of semantic information that affects the topic modeling effect.This paper proposes a user review model UCLDA,which combines user comments and The user's historical Weibo text is integrated to expand the features of the Weibo short text,which alleviates the problem of the sparse Weibo text as short text features.Then the data features are modeled to obtain the distribution of the subject words,and the weighted K-Means calculation method gets Weibo topic clusters.The model uses the microblog data obtained by the crawler as a data set for experimental testing,selects different training sets and test sets according to a certain ratio,and conducts K different experiments.By comparing the clustering algorithms based on UCLDA,LDA,and BTM The experiment verifies that the method based on UCLDA and weighted K-Means has improved the accuracy and effectiveness of hot topic discovery.(2)Aiming at the problem that the topic model cannot be combined with the timeliness of users' preferences on Weibo topics,this paper proposes a fusion similarity algorithm.First determine which of the Weibo topic clusters the new user's Weibo topic corresponds to,and then according to the possibility that the potential attributes of different Weibo content alternate with each other,the similarity is made by naming the user's behavior,Weibo content,and Weibo topic Computational analysis,combined with the influence of the external environment,gives the same weighting factor to the three to calculate their similarity.The model also uses the microblog data set as experimental data,and compares it with the traditional similarity-based recommendation algorithm.The experimental results show that the accuracy of the actual value is significantly lower than the similarity result obtained after fusing the three attributes.This algorithm not only deeply considers the hotspot effect of time,but also improves the influence of hot search topics and unpopular topic effects on the recommendation results.