Event causality extraction through external event knowledge learning and polyhedral word embedding

Extracting causal relations between events from text is vital in natural language processing. Existing methods, which explore the text shallowly, usually aim at casual connection words but neglect implicit causal cues. Furthermore, most of them represent words based solely on contextual semantics, without explicitly considering information related to causality. All of these factors contribute to the inaccuracy of causal relation extraction. To address these issues, in this paper, we propose an event causality extraction method based on external event Knowledge Learning and Polyhedral Word Embedding to alleviate these issues. Specifically, the related background knowledge in knowledge bases is embedded into a vector initially. This infusion of information beyond the sentence allows for the discovery of latent causal relationships. Additionally, we enhance the causal semantic features of words by utilizing their part-of-speech and character features, which helps distinguish causal-related words in sentences. The experimental results on an extended SemEval dataset indicate that our method achieves the best results compared to other existing methods.