XÂY DỰNG MẠNG TƯƠNG TÁC Y HỌC SỬ DỤNG PHƯƠNG PHÁP HỌC MÁY

Abstract

Medical knowledge graphs are an effective tool to describe the interactions and multidimensional relationships among multiple medical factors. Constructing a knowledge graph depends on the size of the graph, which requires analysis and synthesis of a large amount of information. In today's era of data explosion, the above problem becomes challenging to follow with traditional manual analysis methods. Recently, artificial intelligence has shown promising potential to speed up solving big data problems. Following the same approach, we present a building method of medical knowledge graphs applicating machine learning techniques. Specifically, text processing methods screen and selects important medical keywords, incorporating the evaluation function to quantify the interaction between factors. As a result, after analyzing practically 100 medical documents with 76 thousand pages and 32 million words, we filtered out and created a knowledge graph of 438 keywords. Verifying the value of the graph by qualitative analysis of the rationality of the relationship between "symptom – disease" and "symptom - organ" showed a high correlation, compatible with medical knowledge. These preliminary results show the potential of bridging the two fields of data science and medicine, facilitating the acceleration of hidden knowledge extraction in the medical field.