关键词: 社交网络, 微博, 首要事件, 故事线, 图卷积网络

Abstract: As a key platform for users to acquire and disseminate news events, Weibo has rich event information. Extracting storylines from Weibo provides users with an intuitive way to accurately digest event evolution. However, data sparseness and lack of context make it difficult to extract storyline from Weibo data. Therefore, two consecutive tasks for extracting storylines were introduced: 1) Events were modeled by propagation impact of microblogs and primary events were extracted. 2) The heterogeneous event graph was built based on event features, and event graph convolution network model was proposed to improve the learning ability of implicit relationship between events, predict story branches and link events. The proposed method was evaluated from the perspectives of story branch and storyline on real datasets, and the results show that compared with Bayesian Model, Steiner Tree and Story Forest, the F1 value of proposed method is 28%, 20% and 27% higher in dataset-1, and 19%, 12% and 22% higher in dataset-2. In storyline evaluation, the results show that compared with Story Timeline, Steiner Tree and Story Forest, the edge accuracy of proposed method is 33%, 23% and 17% higher in dataset-1, and 12%, 3% and 9% higher in dataset-2.

Key words: social network, Weibo, primary event, storyline, graph convolutional network