Knowledge Graph Construction And Application In Healthy Diet Domain

In the past 40 years,significant change has shown in people's diet structure in developing country with consumption growth of meat product,as a result,the chronic diseases became the main problem of human health.In this background,it is necessary to acquire and retrieve healthy diet knowledge.Fortunately,the Internet has created a search,learning and sharing platform of healthy eating information,but the information is complex and involves many aspects such as food,nutrition,medical treatment,etc.,and is often distributed across multiple data sources,which is difficult for users to search and learn.So how to integrate these resources and extract the knowledge which is implied in semi-structured and unstructured data and provide users and retrieval system with rich structured semantic information that become an important research question.At present,the research of this problem are focused on knowledge graph method and technology.There are many researches and healthcare systems applications based on knowledge graph,healthcare ontology and knowledge base,while most of them are concentrated on the field of professional medicine and the entities and relations are from structured databases and ontologies.In addition,compared with English,the existing Chinese domain knowledge base and training corpus are few,which poses challenges for knowledge integration.According to the characters of healthy food data,this paper defined the main five entities in healthy diet knowledge graph: food materials,dishes,nutrients,symptoms and crowds.Then this paper acquired semi-structured and unstructured text data from multiple healthy websites.After doing entity recognition,relation recognition and semantic disambiguation,a healthy food knowledge graph can be constructed and used for semantic research application.The model and method has good performance in small-scale training set.The primary coverage and contribution of this paper are shown as follows:(1)This paper defined a data scheme of knowledge graph in the field of healthy food,including the types of entities,relationships among entities and attributes.(2)This paper built some domain term dictionaries and designed domain features,then combined them with Conditional Random Field algorithm to apply entity recognition.It can extract symptom,crowd,and nutrient entities in the nature language documents and classify the relationships between these entities and food entities.(3)This paper applied a semantic disambiguation method to solve the problem of the diversity of food names and integrated the same food entities in the knowledge graph.(4)This paper designed a semantic retrieval model proposed for the healthy diet knowledge graph.The results of the experiments showed that the algorithms of entity recognition,relation classification and semantic disambiguation held good precision and recall,which could guarantee the knowledge graph with high quality.Meanwhile,to reveal the application value of this knowledge graph,this work designed a multi-concepts retrieval model and implemented a prototype system,which can effectively manage and retrieve the concepts,entities and semantic relationships in knowledge graph.It is helpful for non-professional users to search and learn healthy food knowledge more efficiently and comprehensively.