Mining Hidden Semantic Information Of Traditional Chinese Medicine Using Topic Models

Mining hidden topics on the documents and extracting hidden meanings of words is essential to many applications in the area of Information Retrieval. We can apply topic models using statistic-learning methods on sample data sets, in order to discover the latent sematic meanings. On one hand, we can improve the accuracy, compared with literal terms matching, on the other hand,"Metaphor Extraction, Comparison, and Classification(取象比类)" is widely used in Traditional Chinese Medicine(TCM) domain. its philosophy consists of using statistic methods to describe the relation between human and nature.LDA Topic model is a three-layer Bayes model, i.e. Document-Topic-Term, it’s an outstanding mathematic model for mining hidden topics. By applying LDA on TCM data sets, we can build the Medicine-Topic-Chemical relation graph. This would be meaningful if we integrating the graph into the current TCM semantic web, and its contribution to modern TCM semantic web may help discover new medicine.This paper contributes on the following aspects:●We present an improved Gibbs-LDA algorithm for topic modeling on TCM data sets.●Using Resource Description Framework (RDF) to state the mined topic knowledge, and presenting a visualization algorithm to make the semantic web graph more accessible to common users.●Integrating the data mining algorithm to the Spora data mining platform, which made it possible for all users to mine their own data.●Designed and realized the Identify and Access Management (IAM) system, which meets high concurrency and could manage billions of resources.