A semi-supervised learning method for Names of Traditional Chinese Prescriptions and Drugs recognition

Knowledge discovery of Ancient Medical Literatures (AMLs) is a research focus due to wide applications of computer technology in Traditional Chinese Medicine (TCM). The foundation of the knowledge discovery research is to get semantic labels within the AMLs and to restructure the text. Due to the diversity of AMLs, low coverage rate of current semantic lexicons and the ambiguities of the lexicon words, low recall rate and low accuracy are resulted by using only lexicons to recognize TCM terms. This paper presents a semi-supervised learning and Bootstrapping based approach, Barpidusk, which aims at using semantic lexicons and simple features to recognize Names of Traditional Chinese Prescriptions and Drugs (NTCPDs) in un-annotated AMLs. And human-computer interaction is added to the Bootstrapping, which increases recognition accuracy without much loss of efficiency. Experiments show that the F values of recognizing Names of Traditional Chinese Prescriptions (NTCPs) and Names of Traditional Chinese Drugs (NTCDs) reaches 44.9% and 51.3% respectively without interaction with humans. By gradually adding human-computer interactions 10 times during recognition process, these values are increased to 74.9% and 90.6% respectively.