The invention discloses a major formula analysis and discovery method, relates to the technical field of traditional Chinese medicine special departments and diseases, doctors and literature major formula big data correlation research, and solves the problems that a correlation rule-based analysis method is relatively fuzzy for definition of correlation among medicines and a clustering-based major formula correlation research method has relatively large difference in eigenvalue extraction and similarity measurement methods in the prior art. According to the major formula analysis and discovery method, each knowledge unit is endowed with coordinate mapping by utilizing a deep learning technology based on a knowledge graph; distance information is fully utilized to reflect correlation among major formulae; the advantages of correlation rules, sample clustering and complex network community discovery can be integrated; and the advantages, namely, multi-dimensional knowledge graph presentation and knowledge reasoning, inexistent in the conventional method are achieved. A user can freely set a community quantity; and the method has remarkable superiority in common medicine pair semantic retrieval, visual traditional Chinese medicine community discovery, single medicine and basic formula correlation research.