Mining Symptom-Herb Patterns from Patient Records Using Tripartite Graph

Abstract EN

Unlike the western medical approach where adrug is prescribed against specific symptoms of patients,traditional Chinese medicine (TCM) treatment has a uniquestep, which is called syndrome differentiation (SD).

It is arguedthat SD is considered as patient classification because priorto the selection of the most appropriate formula from a setof relevant formulae for personalization, a practitioner hasto label a patient belonging to a particular class (syndrome)first.

Hence, to detect the patterns between herbs and symptomsvia syndrome is a challenging problem; finding thesepatterns can help prepare a prescription that contributes tothe efficacy of a treatment.

In order to highlight this uniquetriangular relationship of symptom, syndrome, and herb, wepropose a novel three-step mining approach.

It first startswith the construction of a heterogeneous tripartite informationnetwork, which carries richer information.

The second step isto systematically extract path-based topological features fromthis tripartite network.

Finally, an unsupervised method is usedto learn the best parameters associated with different featuresin deciding the symptom-herb relationships.

Experiments havebeen carried out on four real-world patient records (Insomnia, Diabetes, Infertility, and Tourette syndrome) with comprehensivemeasurements.

Interesting and insightful experimental resultsare noted and discussed.