2016 IEEE First International Conference on Data Science in Cyberspace (DSC) (2016)
June 13, 2016 to June 16, 2016
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/DSC.2016.16
The topic detection of university BBS (Bulletin Board System) plays an important role in studying university students' hottest attention, and it presents the trend of campus opinion. Existing topic model use word probability distribution to represent topic, which lacks of interpretability, and it's difficult to express an unified meaning. What's more, university BBS has its own characteristics and forum posts are colloquial, the existing topic model is unsuitable. Based on the results of LDA (Latent Dirichlet Allocation), this paper proposed a topic label extraction method, including three steps, as topic modeling, keywords extraction and topic selection. Topic select algorithm and artificial feedback mechanism are introduced to improve the accuracy and relevance of topic results. According to the experiment of analyzing the data of BYR BBS, the results show that this method works well.
Hidden Markov models, Dictionaries, Correlation, Data mining, Telecommunications, Probability distribution, Resource management
W. Tang, X. Wu, Y. Li and J. Xu, "A Topic Label Extraction Method for the University BBS," 2016 IEEE First International Conference on Data Science in Cyberspace (DSC), Changsha, China, 2016, pp. 678-682.