| Platforms on the Internet generate a large amount of medical text data every day,such as doctor-patient dialogue data from various medical Q&A websites.The number of medical texts containing rich medical knowledge has skyrocketed.How to efficiently utilize valuable information in texts urgently needs to be addressed.The status of information extraction task,which is the precursor task of semantic understanding,is rising,and it mainly consists of two tasks: named entity recognition and relationship extraction.Named entity recognition and relationship extraction have become the key technologies for medical text information extraction because they can realize the semantic structure of medical text explicitly,so this paper has conducted research on entity relationship extraction for different types of text,but the Chinese named entity recognition task has not yet reached the point of complete solution.The existing character-based Chinese named entity sequence annotation model can reduce the problem of error propagation brought about by word separation,but there are problems of entity boundary and type recognition errors.For entity relationship extraction,due to the ambiguity of entity-entity relationship in a sentence,there may be multiple entity relationships between two different entities,which need to be launched by in-text contextual semantics,and it is difficult for the existing model to solve such problems well.In view of the above problems,this paper studies a Chinese medical named entity recognition model based on local and global character information enhancement and a relationship extraction model based on entity type relevance and entity enhancement from the characteristics of existing deep learning models that can mine deep semantic information and automatically perform information extraction,based on the characteristics of the domain to which the dataset belongs,and carries out the following two aspects of work.(1)Aiming at the problems of strong domain of the corpus for Chinese medical named entity recognition training and insufficient utilization of the potential semantics of characters,this paper proposes a named entity recognition model that integrates the use of local and global character information.The character representation is enhanced by incorporating the local information of Chinese characters’ morphological paraphernalia and the global information of domain terms,which enhances the semantic and potential boundary information of characters and enables the model to obtain better entity recognition ability.We use a self-encoding mechanism to fuse different local information of characters such as spatial information and sequence information of morphological radical,and use an interaction gating mechanism to control the degree of contribution of local and global information of characters to character representation,so as to obtain a comprehensive character representation and finally improve the overall performance of the model.(2)In response to the insufficient utilization of implicit relationships between types in relationship extraction tasks,a method is proposed to enhance character information using entity labels and position embedding,and to enhance local context representation of entities using attention control mechanisms.This article adopts a pipeline approach and achieves better performance in extracting Chinese medical relationships on Chinese texts with sufficient corpus and relatively standardized data.Among them,the entity label in this method not only contains the type information of the entity,but also the position information of the entity in the text and the subject The location embedding further enhances the relative location information of the entities,and then the local contextual representation in the text and between the entities is obtained through the attention mechanism,and the relationship between the subject entity and the object entity is finally inferred by combining the entity type correspondence.The basic idea of this method is to use the attention mechanism to make the model select the most suitable entity type and contextual semantic information for the current context,to improve the performance of the model for entity relationship extraction,and to improve the generalization ability of the model to some extent. |
