Word2Sent: A new learning sentiment‐embedding model with low dimension for sentence level sentiment classification

Abstract

Word embedding models become an increasingly important method that embeds words into a high dimensional space. These models have been widely utilized to extract semantic and syntactic features for sentiment analysis. However, using word embedding models cannot be sufficient for sentiment analysis tasks because they do not contain sentiment features. Therefore, word embedding models do not adequately meet the comprehensive needs of sentiment analysis applications that rely on recognizing the polarity of a sentence. In this paper, we propose a sentiment embedding model (Word2Sent model) to tackle the weaknesses of the existing word embedding models for sentiment analysis applications. We developed this model based on the Continuous Bag‐of‐Words model and SentiWordNet lexicon to learn sentiment embedding for each word from its surrounding context words. It preserves semantic and syntactic features and captures implicitly sentiment ones. Besides, it can predict sentiment features in a very low sentiment embeddings dimension than traditional ones. The proposed method provides an improved sentiment classification performance and lowers the computational complexity. Both the accuracy performance and processing time results obtained indicate that the proposed model is particularly promising.