Causality extraction model based on two-stage GCN

As one of the indirect causality, cascaded causality can be used to construct event knowledge graph, causal inference, scenario analysis, etc. The existing GCN methods lack mining context information and relevant entity information, resulting in poor ability which inevitably affects extraction accuracy cascade causality. To solve this problem, paper proposes a model based on two-stage improve accuracy. obtain rich features entities, work combines sentiment polarity base get candidate library. Firstly, BERT is pre-trained using extracted from library final nodes. Secondly, semantic dependency each possible edge between any two nodes obtained, are input into first stage preliminary directed graph Finally, second achieve deep multi-hop inference. Thus, inferred by model. Experiments show that has been further improved.