Research On Geographic Topic Trend In Online Social Network

Online social network, as the most.active communication platform, has influenced various aspects of the society with its revolutionary power, and brought the social media phenomenon which has drawn great attention. Users in the social network regard it as a self-media for expressing emotions and advocating demands, and public opinions that constantly generated are widely spreading across the media. The highly active and self-organizing mode of information diffusion has made traditional opinion aware techniques limited, thus analysis on macro trends combining with personal information has become a new research direction. This thesis introduces the location information into public opinion aware research. It consists of three parts:the analysis of the semantic and spatial properties in social network, geographic topic detection and geographic topic tracking.The analysis on the semantic and spatial properties of social networks puts emphasis on the comprehensive correlation of users, topics and regions. Firstly, users' regional and topical features are presented based on individual behavior pattern. Secondly, this thesis analyses the effects of locations and topics on the use of terms from writing purpose, document theme and sentences building. In addition, taking the culture regional distribution into account, we abstract the correlation between topics and regions.On the basis of semantic and spatial properties analysis, this thesis introduces the concept of geographic topic and further constructs a probabilistic model called GTDT (Geographic Topic Detection for Twitter). The model combines the user, topic and region into one unified framework, so that it could describe the semantic topic, the spatial feature of the topic at the same time. Experiments on the real social network dataset prove the effectiveness of the model. Meanwhile, compared with Geofolk and LGTA, the proposed model achieves better performance in the criteria of perplexity.Based on the topics extracted by the GTDT, a geographic topic tracking method is designed, in which the topic and document could be represented in one semantic space. The experiments shows that the propoesd method has a good performance.