A Labeled Latent Dirichlet Allocation for Knowledge Extraction from Specific Health Guidance Documents

Abstract

In Specific Health Guidance, professional instructors provide periodic counseling and guidance to those who must change their lifestyle habits as being expected to have or having metabolic syndrome according to the results of checkups. For obtaining useful knowledge of practical guidance from written reports, this study introduces a labeled latent Dirichlet allocation method to develop a topic model of the Specific Health Guidance domain. Individual records of conducted health guidance have labels attached to indicate what kind of advice the instructor gave to his/her clients. By considering these labels, the proposed model estimates the content and distribution of guidance topics, including unlabeled ones. The estimated topic distribution that represents each guidance record in a numeric vector can provide another means to find correlations between guidance aspects and outcome measures.