Semi-Supervised Causal Relation Extraction Based On Domain Adaptation

With the wide application of the Internet and the rapid development of information science,natural language processing technology is facing difficulties while making breakthroughs.The causal relationship extraction task can help computers understand the information in the text,so it is an important task in the field of natural language processing.As a special relationship,causality plays a guiding role in human production and life.A causal network can guide humans to conduct production and life in an orderly manner,and make comprehensive and rapid responses to certain crises.In machine learning,causality extraction can also be applied to question answering,event prediction,intelligent diagnosis,and other fields.At present,research in the field of causal relationship extraction is often based on small hand-labeled corpora,and it is time-consuming and expensive to label causal relationship information from unlabeled data.Integrating causal data from existing public datasets is one available solution.However,these data come from different datasets for different tasks,the data information varies greatly,and the amount is still small.The lack of training data makes the model unable to effectively learn the causal relationship,and the trained model has poor generalization ability and low applicability.To solve this problem in causal relationship extraction research,this study has done two aspects of work.On the one hand,the labeled data is expanded by integrating public data sets;on the other hand,the training method of semi-supervised learning is used to solve the problem of insufficient supervision by using unlabeled data.The research work in this paper has the following three contributions: 1)A semi-supervised causal relation extraction method based on domain adaptation is proposed.The method exploits domain adaptation to align labeled and unlabeled text in the latent space.Experiments show that this method effectively utilizes the information of unlabeled data and improves the performance of models.2)A causal relationship extraction method based on pattern matching and dependency analysis is proposed.Using pattern matching and dependency analysis techniques to extract causal data in an unsupervised manner can be used as a means of data supplementation.3)By augmenting causal data on existing public datasets,the semi-supervised causality dataset for text classification SSCDC and the semi-supervised causality dataset for relation extraction SSCDRE are constructed,which provide more data support for research on causality extraction.