Using a knowledge graph for hypernymy detection between Chinese symptoms

Hypernymy relationship plays a critical role in language understanding because it enables generalization, which lies at the core of human cognition. As an important part of medical text understanding, Chinese symptom hypernymy detection is important for some medical applications, such as intelligent diagnosis, drug mining, and similar medical record analysis. However, there exists the problems that some symptoms are not included in the training data, and some pairs of symptoms do not appear in the same sentences when path-based methods are used. In light of that, we propose a new method which use a knowledge graph to detect Chinese symptom hypernymy relationship. Specifically, we first construct a symptom component knowledge graph, including atom-symptoms, body parts, headwords, and their synonyms and hypernyms. Then we use the constructed knowledge graph to standardize symptom pairs and detect hypernymy relationship between them. Experiments show that our method achieves state-of-the-art performance, with 98.90%, 87.25%, and 94.34% in Precision, Recall, and F-score, respectively.