Topic Tracking Text-based Network Public Opinion Research

With the rapid development of network, today's society has entered an era of explosive information, people are free to use the network to express their views and attitudes on certain things, the network has been gradually become an important birthplace of public opinion that spreads. Flooded every corner of the Internet for information that is not conducive to people's lives and harmonious development of society, caused a great deal of threat to social and public security, so researching the public opinion of the network has far-reaching significance in maintaining social stability and promoting national development.Topic Tracking is a sub-task of topic detection and tracking, and its purpose is to monitor the flow of news reports, identify follow-up reports related to the known topic. The main work of this paper is to study the text-based Internet public opinion of topic tracking.The main works of the thesis are as follows:1. Researching the related and key technologies of topic tracking:web crawler, text extraction, Chinese word segmentation, feature selection of news reports, the weight calculation, topic or report model's building and the calculation of the similarity between the two models;2. Implementing the algorithm of adaptive topic tracking based on the topic update. In order to resolve the problem of training report sparse of the traditional topic tracking algorithm and the topic drift of adaptive topic tracking algorithm, this algorithm uses the idea of adaptive information filtering, and updates the topic model,so that it improves the fitness of the topic model;3. Proposing two temporal information-based adaptive topic tracking algorithm:an adaptive topic tracking algorithm with similarity adjustment based on the temporal information, an adaptive topic tracking algorithm with dynamic threshold based on the temporal information, and giving the processes. Both of these two algorithms use the advantages of adaptive topic tracking algorithm based on the topic updated, and according to the dynamic characteristics of the news reports over time, using temporal information, considering the issue from two different perspectives—similarity and threshold, proposing two algorithms. Experiments show that these two algorithms have good performance;4. Presenting the overall design of a topic tracking system which based on text and Internet public opinion, and giving the realization of various sub-modules of the system;5. Proposing the experimental methods is to determine the best number of feature selection and best initial threshold. And experiments prove that select the best number of features can reduce the number of feature vector dimension, but it can make the system performance has stabilized, and it can also prove that the best initial threshold allows the algorithm to achieve optimal performance.