Evaluating Drug Entity Relationship Using Word Embedding as Feature

Abstract

In this paper, a method is proposed to evaluate the Drug Entities Relationships (DER) using word embedding as a feature. Machine learning models are employed to produce word embeddings and relationships are intrinsically evaluated by measuring the similarity between word embeddings. This methodology is split into two directions. One direction will focus on embedding dimension and window size which effects the semantic information of one entity with other entities in vocabulary. The other direction will focus on the potency of similarity indices, cosine distance and Euclidean distance used to measure entities similarity. Our results indicates the effectiveness of measuring entities similarities for evaluating semantic relationships on the corpus built from the DrugBank database.