Sentiment Analysis through Word Vectors: A Study on Movie Reviews from IMDb

The present study emphasizes the challenges of labeling the correct semantic orientation of movie review texts due to the high degree of ambiguity and varied narratives of the reviewers. The study uses a vector-based model of sentiment analysis on IMDb movie review data set. The model computes the attributes of text reviews based on singular words and the longest common subsequence of words. We present a comparative assessment of unsupervised vector-based model (K-means clustering) and supervised models (support vector machine with linear kernel and naive Bayesian classifier) for sentiment-based classification of movie review data set. The study reports that supervised models are outperforming unsupervised model in terms of accuracy. Despite the high degree of ambiguity of movie reviews, the results are encouraging in the context of natural language processing.