Refining Word Embeddings Using Domain-Specific Knowledge for Drug Reviews Sentiment Classification

Abstract

With the development of internet technologies, more users are sharing and seeking health-related information in online medical forums about medications and treatments and their contents are usually in subjective nature. Patient-generated drug reviews are valuable and useful textual contents which have not been researched largely by researchers in the natural language processing area. Analyzing drug reviews can assist to improve the pharmacovigilance systems. We propose a refining technique for traditional word embedding using domain-specific knowledge for the drug reviews sentiment classification. The domain-specific lexicon generated from the drug reviews corpus is applied to refine traditional word embeddings in the feature extraction process of sentiment classification. We evaluate our proposed method on the publicly available drug review datasets. According to the experimental results, the proposed method outperformed refined word embeddings using domain-independent lexicon in terms of accuracy in sentiment classification of drug reviews. It indicates the significance of domain knowledge in sentiment analysis of medical domain.