In recent years, more and more studies have been done on symptom information extraction.
These studies are mostly based on clinical medical records, and they focus only on symptom
entities, which are not sufficient to convey the full symptom information. This paper presents a
feature split-based approach to extract symptom information from Chinese medicine instruction
texts. In this approach, the symptom information is split into two parts: symptom subject entity
and symptom manifestation entity. The main idea of this method is to automatically recognize
the symptom subject and symptom manifestation first, and then add these two identification
results as features to the symptom information extraction task. Through a series of experiments
based on Conditional Random Fields (CRF)- an effective model proved by lots of experiments in
the field of medicine, it is obvious that the feature split-based approach proposed in this paper
can obtain higher accuracy and recall rate in symptom information extraction.