A Topic Extraction Method for Network Public Opinion

Network public opinion has played a vital role during the government's decision-making process. There have been many existing topic extraction methods on processing network public opinion, while most of them have paid attention to its short text characteristics and have underutilized its evolutionary characteristics over time. This paper intends to hybrid the textual and evolutionary characteristics during the topic extraction and proposes a phase-based topic extraction (P-TE) method. Firstly, a novel idea about combining qualitative and quantitative methods is developed, and the evolutionary characteristics are used to divide the process into several phases by time series analysis. The number of phases depends on the certain event and differs from each other. Then, based on the textual characteristics of network public opinion, feature extraction is described. Finally, topics are extracted for every single phase separately. The experimental results show that P-TE can reveal more details and careful thoughts about the event than other methods. Furthermore, the rationality of P-TE is verified.