Research On Knowledge Graph Construction Of Online Medical Community Q&A Texts

With the rapid development of the Internet and the advent of the Web2.0 era,people have more and more ways to acquire knowledge.Today's fast-paced life makes people pay more attention to medical and health problems.Compared with the traditional way of consulting doctors in hospitals,people are gradually accustomed to obtaining relevant medical information through the online medical community to solve some simple health issues of themselves or their families,while improving awareness and prevention of related diseases.Thousands of text records have been accumulated in the question-and-answer sections of the medical community with a certain user scale such as "Good Doctor Online","Xunyiwenyao",and "39 Health Website".The format of the text has the characteristics of a community text with large data volume,poor standardization and sparse data.It also has the medical text characteristics of professionalism and complexity,and has good research value.Knowledge Graph as a way of knowledge representation is essentially a semantic network technology.In the medical field,the research of applied knowledge graph is mostly focused on Electronic Medical Record(EMR).There is not much research on online medical community Q & A texts.Therefore,based on the Q & A texts of the medical community,a comprehensive use of Bidirectional Long Short Term Memory(BiLSTM),Conditional Random Field(CRF),and Bidirectional gated recurrent neural network(BiGRU)And Attention deep learning model,successfully constructed an online medical community breast cancer knowledge graph.First of all,this article selects the breast cancer section of Xunyiwenyao as the research object,crawling a total of 12626 pieces of question and answer data on the page of this section;performing simple preprocessing on the data by cutting words and segmenting words and removing stop words;Then use Word2 vec to pre-train the word vectors;then perform entity recognition and relationship extraction.During the entity recognition process,the word cloud(Word Cloud)is used to further classify the entities objectively,and then use the BiLSTM-CRF model to label the BIO data Set for entity recognition,the experiment found that the subdivided entity performed better than the unsubdivided entity in results;then the BiGRU-Attention model was used to extract the relationship between the entities.The experimental results show that whether the model is in accuracy,recall or The F value is greatly improved than the BiLSTM-Attention extraction model;afterwards,entity recognition operations are performed on all recognized entities;finally,the Neo4 j graph database is used to construct a visual knowledge graph and analyze it from the display level and management application level.Research summary and contribution: This study converted unstructured community text into structured data,and successfully constructed an online community medical Q & A text knowledge graph.Second,the first time to use the word cloud tool to objectively subdivide entity annotation types,making Entity recognition is more accurate;Finally,the online medical community Q & A text knowledge graph constructed in this paper has a promoting role in the medical community's intelligent knowledge services,knowledge representation,personalized knowledge recommendation,etc.,providing personalized medical and other online community research A new direction and service ideas have been introduced.