Semantic Relation Extraction for Herb-Drug Interactions from the Biomedical Literature Using an Unsupervised Learning Approach

Abstract

Sharing principles of drug-drug interaction, herb-drug interaction (HDI) investigates the impacts of herb-based products on activities of other conventional drugs when combining them in certain medical treatments. For years, patients using herb-based medications have built a misconception about the absolute safety of products derived from natural sources. The current fact revealed that patients had intentionally combined herb-based products and prescription drugs for any certain illnesses without safety concerns to enhance the efficiencies. Incapability of non-experts in reviewing the biomedical literature of potential HDIs may be considered as one of the most reasonable explanations for this issue. In this study, text mining techniques are applied to provide users with a novel approach to save time when looking for information of HDIs. Since constructing an annotated corpus for herb-based products in traditional manner requires a high demand for human resources and financial support, an unsupervised learning model for relation extraction which eliminates to the crucial role of an annotated training set is quite suitable. The relations connecting the entity pairs were discovered and labeled by their most significant features. The obtained result proposes a promising method for the HDIs extraction challenge.