Event Evolution Analysis of Network Text Based on Pre-trained Language Model and Event Graph

Event evolution analysis can help understand the current status and trends of our society, has attracted a lot of interest from the industry and academia. Previous evolution methods include the storyline-based methods and event knowledge graph-based methods. However, the existing methods have many deficiencies. We propose a framework of event evolution analysis for network texts, which includes a novel event extraction model, an event relation prediction model and a new type of event knowledge graph. Our event extraction model outperforms the previous best model by 2.9% in trigger extraction F1 score, and 3.37% in argument extraction F1 score. Our approach to exploring event relations and connections simplifies the traditional event relation extraction method, and is experimentally proven effective.