Tracking Based On Adaptive Topic Microblogging Development Time

With the rapid development of the Internet, social networks are favored by more and morepeople because of its interactivity, freedom and openness. At present, the Internet is to meet thedemand of people toward the development of interactivity and openness. In2006, the firstmicro-blog service---Twitter in world is launched by the company of Obvious, and since then,the micro-blog service is coming into people’s horizons quickly. Unlike conventional news andblogs, the content of micro-blog is brief, and limited to140words. However, except the brieftext, the user can add pictures, video, audio and other links in their micro-blog content. Themode of transmission with freedom and openness, is welcomed and noticed widely by themajority of users, and an upsurge of micro-blog services is pushed by the users.As the freedom, interactivity and openness of micro-blog, people can share their experienceor express their emotional and attitude at anytime and anywhere. With the rapid growth ofmicro-blog users, some emergencies are easily showed up in the micro-blog platform. Therefore,topic detection on micro-blog is becoming a hotspot of research. However, sometimes, the statusof development of an event is more concerned about by people, so the topic tracking onmicro-blog is particularly important. In order to make full use of the time sensitivity ofmicro-blog, detects and tracks the hot topics on micro-blog timely, this paper conducted theresearch as follows:1. The micro-blog is large amount of information and time sensitivity, for this problem,a method of hot topics found based on speed growth was proposed.In this paper, a method of hot topics found based on speed growth was presented. Firstly,the pretreated micro-blogs were divided on the basis of the equal number of window, andadded up the term frequency in each window,and expressed as feature trajectory of binary groupsequence; Secondly, calculated the growth slope of every adjacent two windows to find thewords with growth speed; thirdly, calculated the growth speed of the word’s relevant users andthe growth speed of the word’s relevant micro-blogs to ensure the word is hot subject or not;Finally, hot topics were found through the hot subject clustering. The result shows that themethod has a strong ability of hot topics mining.2. A phenomenon of topic drift was often showed up due to the evolution of the topic, amethod of adaptive topic tracking on micro-blog based on development of the time wasproposed.Firstly, in topic tracking on micro-blog, a problem of data sparseness was often showed up.For this problem, using a method of feature expansion based on relevant retrieval to expand thefeatures; secondly, a phenomenon that the weight of feature words is constant which lead to thelow recall rate, for this problem, using a method of weight adjustment based on time decay todecay the weight of feature word when we adjusted the topic model; lastly, a phenomenon oftopic drift was often showed up due to the evolution of the topic, for this problem, a method oftopic model updating based on double filtering technology was proposed, the micro-blog whowas determined to be relevant story and the importance score was high was used to update topicmodel. 3. Designed and implemented the network public opinion monitoring system based onadaptive topic tracking algorithm.The method based on adaptive topic tracking was used on the topic model updating in thenetwork public opinion monitoring system. The mico-blog whose importance score was highwas used to update topic model to make the system has higher recall rate and accuracy rate,andmeet the needs of users.