Constructing sentiment sensitive vectors for word polarity classification

Sentiment classification has been an essential part of opinion mining and sentiment analysis. This topic has been applied to real world scenarios such as mining customer reviews on merchandise sold online and film reviews of movies. Therefore, we aimed to gain insight into sentiment word classification, as it could serve as the foundation for larger scale sentiment analyses on corpuses and documents. In this paper, we focus on word polarity classification, which could be extended to perform classification of sentences and paragraphs. We enhanced our previous work on gloss vector and expanded it with a more concise method to generate the vectors. Additionally, we used more sources to validate the similarities of the candidates with two vectors, each representing the positive and negative sentiment polarity respectively by importing groups of words that express that polarity. Experiment results demonstrated that our method is effective, while producing better accuracies than the previous attempt on similar subjects.