- 期刊
Causality Extraction of Fused Character Features with BiGRU-Attention-CRF
摘要
As an essential sub‐task in natural language processing, rich word vector representation is beneficial to the performance of model causality extraction. We propose a causality extraction method based on IDCNNs for extracting character features and BiGRU‐Attention‐CRF. This method converts causal extraction into a sequence labeling problem by first selecting character features using IDCNNs and merging them with contextual string embeddings and pre‐trained word vectors to form a feature vector to enrich word representations from different granularities, then inputting this feature vector into the BiGRU‐Attention‐CRF network to acquire contextual representations and get the more significant causal features. Finally, the weighted characteristics are passed through CRF to obtain the optimal labels. The experimental results show that the method achieves an 81.06% value on the SemEval 2010 task 8 dataset and improves the value by 1.73% compared to the CNN method for extracting character features, which proves that the model can effectively improve the accuracy of causality extraction.