A method of instance learning based on finite-state automaton and its application on TCM medical cases

In traditional Chinese medicine (TCM) field, medical cases are viewed as semi-structured text, which is between free text and structured text. They lack of grammar, have no strict formats, and even don´t have complete sentences. Most of them consist of phrases having the characteristics of TCM field. Presently, the information in TCM medical cases is extracted based on structured templates. This process requires the experts to take part in. Moreover, each of the experts has their own characteristics. If we use uniform templates to describe the TCM medical cases, they will not only result in the loss of some information, but also not reflect each expert´s idea perfectly. In this paper, a method of instance learning based on finite-state automaton is proposed, after analyzing the characteristics of TCM medical case´s structures. This paper presents a method to automatically generate extraction structure patterns of symptom phrases by instance learning. These structure patterns are expressed by finite-state automaton. By using this method, information can be extracted from TCM medical cases automatically, and the state transition diagram can be used in the traditional Chinese medicine domain to standardize the symptom information phrases. Moreover, information in TCM medical cases is not lost, and each expert´s idea is reflected more perfectly.