Design And Application Of Semantic Recognition Model Based On Diabetes Knowledge Graph

With the rapid development of China's economy in recent years,people are paying more and more attention to their health,and the widespread popularity of the Internet has also greatly promoted the development of Internet healthcare.At the same time,the Internet has also accumulated a large amount of industry data during the development process of many years,including medical-related scientific and technological papers,hospital medical orders,and digital medical data.Numerous related data resources are distributed on different platforms,forming data islands.When people search for related information,they need to sort out relevant information from different search results by themselves,which will cost people a lot of energy.The knowledge graph concept proposed by Google for intelligent search can be used to integrate heterogeneous data from different sources.This article will use the technology of knowledge graph to build a diabetes-related knowledge graph.On the one hand,it hopes to help scientific research or medical practitioners can easily query related knowledge,on the other hand,it also hopes to provide data support for various intelligent services related to diabetes.The main research contents of this article are as follows:First,the overall construction process of the diabetes knowledge graph is designed.The knowledge graph contains a total of 15 entity categories and 10 relationship categories.The Arango DB database is used to store the knowledge graph,which can provide data-level support for various intelligent services.Secondly,in the process of constructing the diabetes knowledge graph,for the entity recognition task,this paper introduces multi-level word vector features based on commonly used algorithm models.The experimental results show that the method introduced multi-level word vector and character vector features in this paper is slightly better than the previous basic models.In order to identify the relationship between entities,this paper first conducts experiments on two basic models and incorporates the idea of multi-level word vector features into the models.In addition,this paper proposes to introduce entity category information in the process of relationship recognition.After a series of comparative experiments,it is shown that the feature of the entity's category information can significantly improve the accuracy of the model,and its accuracy is improved by 6%-9%.Among them,the model with the highest experimental scores uses the character features of textual information,multi-level word vector features,entity location features,and entity category features.Finally,based on the constructed diabetes knowledge graph,this paper proposes a semantic recognition model with two tasks,identifying intent of the user's search text and key entities in the sentence.Based on these two purposes,the thought and construction process of pipline semantic recognition model and joint semantic recognition model are described respectively.The pipline semantic recognition model first performs entity recognition,the method used is similar to the algorithm for the entity recognition task of the knowledge graph construction process.And then the intent recognition is performed,the model is used to indicate what the central topic the user ultimately wants to search.Last,the result of entity recognition and the result of intent recognition are checked for knowledge consistency.The joint semantic recognition model is a combination of the two tasks described above in one model.It requires the weighted summation of the loss function of entity recognition and the loss function of intent recognition to continuously modify the parameters of the model,and the joint model uses the idea of N-best,by performing a Cartesian product operation on the results of entity recognition and the results of intent recognition,and reordering all results of Cartesian products,and using the sorted results as the final output of the model,ensuring the consistency of entity recognition results and intent recognition results.This article also illustrates two application scenarios of the semantic recognition model in semantic search and shows the results.