Article Preview
Top1. Introduction
Sentiment analysis is becoming a trending and popular amongst researchers. Sentiment analysis is a special case of text classification which aims to categorize opinions based on polarities for e.g. positive or negative. People can express their opinions on the web in the form of reviews. The reviews can be extracted from different fields such as product review, movie review, tweets from Twitter, etc. Basically, sentiment analysis is done to express what a person thinks for a particular product, movie, twitter etc.
In recent years lots of people are expressing their opinions on the web. Everyday millions of people express what they think in the form of reviews on platform such as blogs, forums, social networking sites like twitter etc. These platforms help people to connect to other people whom they don’t even know and get the opinion from them. Basically, these platforms serve as intermediate between the end user and the service provider. From the end users perspective, these platforms are useful to get an idea about the product. From the service provider point of view, it helps to improve the standard of their product and services.
The data on the web is increasing everyday as many people are using it as a platform to express their reviews and opinions. Therefore, it becomes a very hectic task for people to take instant decision about the reviews and opinions. People are confused that on which review they can trust and on which review they cannot. Due to this problem, an automatic system must be designed to make the process of analysis, summarization and classification easy. The general approach for sentiment analysis is the Bag of Word (BOW) Approach (Dave, Lawrence, & Pennock, 2003). In this approach, a document is divided into bag of words to make feature vector which will be used in the classification process. However, BOW approach is failed to generate the desired results as it doesn’t capture word sequence and semantic relation. The works have been carried out in the field of sentiment analysis to improve BOW in combination with linguistic knowledge (Dave, et al., 2003; Gamon, 2004; Kennedy & Inkpen, 2006; Na, Sui, Khoo, Chan, & Zhou, 2004; Ng, Dasgupta, & Arifin, 2006; Pang, Lee, & Vaithyanathan, 2002; Whitelaw, Garg, & Argamon, 2005; Xia, Zong, & Li, 2011). However, they failed to improve the classification accuracy.
The polarity shift is an important issue in sentiment analysis. Many approaches have been suggested in the literature to overcome the polarity shift problem (Councill, McDonald, & Velikovich, 2010; Das & Chen, 2001; Ikeda, Takamura, Ratinov, & Okumura, 2008; Li & Huang, 2009; Li, Lee, Chen, Huang, & Zhou, 2010; Wilson, Wiebe, & Hoffmann, 2009). However, most of them required either complex linguistic knowledge or extra human annotations. Such high-level dependency on external resources makes the system difficult to be widely used in practice.
Sentiment Analysis can be carried out at different level such as aspect level, sentence level and document level. The proposed method uses sentiment analysis at sentence level using the lexical resources SentiWordNet3.0 (Baccianella, Esuli, & Sebastiani, 2010) and Affin111 (Zol & Mulay, 2015) to decide the overall polarity of the reviews based on feature vector and machine learning algorithms.