Research And Implementation Of Medical Knowledge Graph Q&A System Based On Deep Learning

As the Internet grows,so does the number of people who seek medical knowledge online.Meanwhile,various online platforms such as medical knowledge websites and health assistants has appeared,which have increased the ways for people to learn about health.However,while these platforms bring convenience to people,there are also many challenges,for example,people has not utilized the massive medical data effectively,the quality of platforms varies,and the feedback of consultation platforms is too slow.Therefore,to deal with the above problems,this paper investigates the entity and relationship extraction model for medical data and further designs and develops a Q&A system based on medical knowledge graph,which performs medical entity recognition and question intent classification according to user input questions,and then returns the most accurate answer to query.The main research content of this paper is as follows.(1)To address the problems of poor language comprehension,insufficient utilization of entity information,and weak extraction ability in complex contexts in existing entity-relationship extraction models,we introduced the Ro BERTa-wwm model and gated attention units to extract text features.We used attention mechanisms and pointer networks to construct an entity-relationship extraction model for Chinese medical texts,which effectively improves the model’s performance on multi-relationship texts and the model can effectively improve the triadic extraction ability on multi-relational text and text with overlapping relations.The model first performs text vectorization by a Ro BERTa-wwm pre-training model,feature extraction by gated attention units,and next constructs a joint extraction model by using a pointer network to first extract head entities and then perform relationship and tail entity recognition.(2)To more effectively use textual information for entity recognition in the phase of user problem detection,this paper proposes a named entity recognition method using dual-channel feature extraction.The method uses Bi-GRU to obtain contextual information while using IDCNN to obtain contextual local features of text.In addition,the model introduces a multi-headed attention mechanism to obtain character-level relationships,which can better identify the target entities in the problem.In the intention recognition stage,this paper investigates the intent recognition method based on Ro BERTa-wwm-Text RCNN.The method uses the Ro BERTa-wwm pre-training model for the question text vectorization,followed by feature extraction using Text RCNN to obtain the target relationships in the problem text.(3)Based on the technology of entity and relationship extraction method,entity recognition and intention recognition method proposed in this paper,a Q&A system based on medical knowledge graph is constructed.And the system realizes the functions of medical Q&A and knowledge retrieval,which makes it more convenient for users to obtain medical knowledge and make self-diagnosis for some basic diseases in a more convenient way.