Detection And Trend Prediction Research Of Hot Topic Of Micro-Blogging

With the innovative production and dissemination mechanisms, microblogging hasplayed an important role of the birth and transmission of hotspot. After analysis of thepropagation characteristics of the hot topics, this thesis proposed a detection method of hottopics base on Opinion Leaders, and put forward a trend prediction method of hot topics.Hot topic detection consists of several steps, first writting some crawlers to collectinformation of opinion leaders and microbloggings, and then characterizing the data andpreprocessing texts. The third step is building VSM(vector space model) of microbloggingtext, and calculating the semantic similarity between texts. After that, sorting themicrobloggings base on the heat, and using Single-Pass clustering algorithm to extract thebasic topic sequence. The last step is identifying the hot topics from basic topics base on thethreshold. To predict the trend of hot topics, using the microblogging Opinion Leadersparticipation rate, the retweeting rate and commenting rate as the influence index, and usingthe number of microblogging associated with the topic in unit of time as the developmenttrend index, building a multiple linear regression prediction model based on the dynamiccoefficient. At last, this thesis tested the results of hot topics and the performance of thealgorithm, and had a summary of the result of prediction in the end.The contribution of this paper is mainly reflected in three aspects:1. Integrated a variety of data collection methods, break through the the microblogging APIfrequency limit within a certain range, implemented simulating search of microblog websit.2. Proposed a more simple detection method based on the topic of microblogging OpinionLeaders,and avoiding the massive data mining of topic. In addition, accuracy and efficiencyof this method had been greatly improved to the randomly selected microblogging method.3. Built a dynamic coefficient of multiple linear regression prediction model, which canpredict the trend of the same type of different topics.