关键词: 知识关联挖掘, 领域知识分析, 预训练, 表示学习, 小规模语料

Abstract: In order to solve the problem that the results of knowledge association mining do not need to be mined in subdivision domain, this paper proposes a new scheme of non-common knowledge association mining which is more suitable for the needs of professionals.The scheme consists of three key points: firstly,to ensure the mining results can better solve professional problems, professional experience-sharing texts are used to analyze data, not common-sense encyclopedic texts. Secondly, to solve the problem of insufficient corpus in subdivided field and the unlisted terminology, the method of "large-scale pre-training word vector and small-scale subdivided field corpus learning fine-tuning" is adopted and carry out better representation learning in domain terminology. Finally, potential and cued knowledge association is provided to professionals after eliminating common knowledge association from mining results. Taking cardiovascular field as an example, knowledge association that mine from experience exchange texts of small-scale can better fit difficult clinical problem solving experience, medical research experiment discovery, and provide valuable clues for professionals to further explore and utilize knowledge.

Key words: Knowledge association mining, Domain knowledge analysis, Pre-training, Representation learning, Small-scale corpus