Harnessing the power of hashtags: Temporal pattern mining and storyline construction for event evolution on social media

With the rapid development and wide application of social media, Weibo, as one of the major social media platforms in China, plays an important role in connecting users with information. However, the huge amount of Weibo data poses challenges for effective analysis and understanding. Timeline construction is critical for understanding event progression, enabling stakeholders to track public opinion shifts, identify critical phases of event development, and formulate timely interventions. This paper proposes a framework to systematically model event evolution by analyzing temporal patterns and semantic correlations of hashtags. We first adopt a temporal feature extraction method to capture the temporal information of Weibo posting time. Then, the correlation between topic tags is considered comprehensively by combining the temporal information and the similarity calculation method of topic tags. Finally, a timeline-based topic merging algorithm is proposed to construct a clear and orderly event story line. Meanwhile, this paper also introduces the RoMLP-AttNet model, which significantly improves the classification recall and precision in the Weibo event classification task by using the topic posting sequences as the background data for assisting event detection. Using the "Japanese nuclear effluent" event as an example, the story line construction method proposed in this paper generated a clear and complete story line. Experimental results demonstrate that the RoMLP-AttNet model proposed in this paper achieved an average increase of 16.73% in recall rate, 15.8% in precision, and 17.38% in F1 score.