A Preliminary Work on Symptom Name Recognition from Free-Text Clinical Records of Traditional Chinese Medicine using Conditional Random Fields and Reasonable Features

A preliminary work on symptom name recognition from free-text clinical records (FCRs) of traditional Chinese medicine (TCM) is depicted in this paper. This problem is viewed as labeling each character in FCRs of TCM with a pre-defined tag (“B-SYC”, “I-SYC” or “OSYC”) to indicate the character’s role (a beginning, inside or outside part of a symptom name). The task is handled by Conditional Random Fields (CRFs) based on two types of features. The symptom name recognition FMeasure can reach up to 62.829% with recognition rate 93.403% and recognition error rate 52.665% under our experiment settings. The feasibility and effectiveness of the methods and reasonable features are verified, and several interesting and helpful results are shown. A detailed analysis for recognizing symptom names from FCRs of TCM is presented through analyzing labeling results of CRFs.