The Study Of Module Network And Community Mining Based On Topic Model

With the development of computer and network technology, a large number of digital libraries were built to provide us with a rich of literature data and digital information resources. Literature data, including a lot of information, such as text data, link information, social information, through the analysis and mining on these data, there can be found a large quantity of useful or potential information, which can improve the researchers to collecting and summering the scientific literatures, understanding the research field, and helping or guiding researchers to do the scientific research work more effectively.This paper has made the following literature data oriented research works:1) With the study on some classical probabilistic generative topic models, we used the conference information in the literature data, design a new topic model called Conference-Author-Topic model. With this model we can capture not only the topic information of the whole literature data, but also the topic distribution of the authors and the conferences. It both make the topic model more accurate and reasonable, but also facilitate the further analysis and research on the author or the conference;2) With the topics extracted from the topic model, we provide a new method to construct the topics' mutual influence modular network. First we get the strength of time-series data of the topics from the result of topic model, and then applied the linear piece-wise segmentation to eliminate fluctuations of the data, finally construct the module network according to the network modularization algorithm.3) Different from the traditional method of community mining, we provide a new community mining method based on the topic extraction. Face to the network of researchers, we could using the topic information, which can reflect the nature relationship of the community, directly to mining the community that has common research interests. It is true of the practical needs and community sense.4) With the studying of the structure composed by the community and its members, we designed a PageRank-style algorithm to rank the communities. It is different from the traditional evaluation method that is always based on evaluating some statistics variables of the community.