Causal Relation Extraction Between Biomedical Entities

In recent years,with the innovation of biomedical experimental methods,relevant experimental data and literature resources have grown exponentially,therefore how to extract valuable information quickly and efficiently from such large-scale resource is a challenging problem to be solved.Causal relation extraction between biomedical entities is a task that automatically extracts causal relationship between entities and their associated functions from biomedical literature.This article conducts research on this task with its main content as follows:(1)Instance-level causal relation extraction corpus construction based on word alignment.Since the original training corpus annotated at sentence level,it cannot be directly used for traditional machine learning.This article automatically constructes an instance-level hierarchical sequence labeling corpus via word alignment from a relation-sentence and word-sentence parallel corpus,laying the foundation for extracting causal relations between entities by using machine learning methods.(2)Causal relation extraction based on hierarchical sequence labeling.Based on the instance-level training corpus constructed above,a hierarchical sequence labeling model is trained with conditional random field,which can further be used for extracting hierarchical causal relations.The experiments demonstrate that the hierarchical sequence labeling model based on conditional random field can perform well on the causal relation extraction task.(3)Causal relation extraction based on hybrid model.Considering the complex relation patterns,the sequence labeling model cannot fully capture the patterns of this level.This article proposes a hybrid model which combines sequence labeling model and classification model,that is,use the sequence labeling model to recognize functions and classification model to identify relationships respectively.The experiments indicate that the method based on the hybrid model can achieve better performance on causal relation extraction.