Sentiment Analysis and Corpus: Cognitive Perspective and Overhead-accuracy Tradeoff

Abstract

Human logical thinking is in the form of natural language. With the development of computer science techniques, it becomes easier and more convenient for natural language processing. Therefore, various natural language processing applications have emerged. Sentiment analysis is one of these novel applications and has been applied in many areas. In Amazon.com, there are a large number of user comments and product discussions, which can help a person to decide whether to buy a product or not without asking the opinions from friends and family members. Therefore, sentiment analysis on user comments and product discussions, such as Amazon reviews, becomes increasingly useful and important. In this paper, the effect of corpus on sentiment analysis of the Amazon review dataset with the aid of support vector machine is studied. We generate eight different size datasets from the Amazon review dataset filtered by different word frequency in Corpus of Contemporary American and conduct some experiments on these eight datasets. According to the experimental result, we make some conclusions and give some suggestions to facilitate researchers to make a trade-off between accuracy and experimental cost.