Improving the Accuracy of Sentiment Classification using Optimized Word Vector

Abstract

Sentiment Analysis is emerging research area under Natural Language Processing (NLP). Sentiment Analysis has wide range of applications such as finding Customer opinion, digital marketing, politics etc., NLP tries to understand the meaning of the word from human perspective. Developing a Sentiment Classification model is a challenging task, as it involves lot of computation and memory requirement is high. Even then number of researchers has tried to simplify the model construction. There are numerous methods to capture the Sentiment from corpus, in which word embedding is showing better results. The word embedding models such as word2vec and Glove are widely used today. The disadvantage of the word embedding model does not perform well on small corpus. The proposed optimized word vector model improves the accuracy of the word embedding models. In our approach, we create vectors on the results of traditional Lexicon, POS tagging, word position algorithm and concatenate those vectors. The result shows that optimized word vectors has improved the accuracy of Sentiment classification.