Causal Relation Extraction Of Biomedical Entities Based On Deep Learning

Biomedical entity relation extraction is an important part of biomedical text mining,which plays a key role in the establishment and application of biomedical knowledge base.Among them,the extraction of causal relationships aims to extract the causal relationships between entities from biomedical literature,and convert them to structured representation.This paper adopts deep learning methods to conduct research on the task of causality extraction between biomedical entities.The main research aspects are as follows:First,this paper proposes a causal relation extraction approach combining relation extraction and function detection,which decomposes the task into two subtasks:entity relation extraction and entity function detection.Two attention-based Bidirectional Long Short-Term Memory Networks(BiLSTM)are applied to extract entity relation and entity function respectively,then the entity relation and their functions are combined into a structured form In addition,the strategy of threshold filtering is used to improve the precision of detected entity functions in order to boost the overall performance.The results show that the approach achieves better performance on the task.Second,the paper proposes a joint learning model sharing decision-making by both binary relation extraction and unary function detection.This method mainly captures the connection between two subtasks by information interaction,in order to improve the precision of function detection task without post-processing and enable function detection to make positive contribution to statement extraction.In the joint model,two subtasks share the embedding representation then BiLSTM models and gated mechanism are used to capture the interactive representation between two subtasks,finally predictions are made respectively under the interactive representation.Empirical studies indicate that the joint learning approach can improve the performance of causality extraction compared to the pipeline way.Finally,the paper presents a method of causal relation extraction based on corpus transfer,which uses the source GENIA event corpus to augment entity function detection,in order to improve the overall performance.Events in GENIA corpus are first converted into functions in the causality extraction task by mapping huristics to form a GENIA function corpus.Next,it is used as a training corpus for entity function detection,via direct classification or event trigger recognition respectively.Last,the extracted entity functions and relations are merged into a structured form The experiment results demonstrate that the approach can improve the performance of causality extraction.