Customer Sentiment Classification and Rating System Based on Products Reviews

In the present competitive business scenario vast amount of consumer reviews are written on Web about any product or service. WWW contains an overwhelming volume of customer reviews about different categories of commodities avail. The advent of social media and ecommerce has brought the era of a new age business and its customer base is growing exponentially every year. In today’s world, the online market is increasingly getting popular and it becomes more and more important to help the customer get the best product by all parameters. The quality of a product is best confirmed by taking the customer reviews from those who are already using that. All popular shopping websites like Amazon, flipkart, ebay etc. allow customer reviews once the product has been purchased. These reviews are such huge in numbers on these websites that it is not possible for a customer to consider them all. This paper focuses on extracting the features from product reviews taken from amazon.com or ebay.in sites given by reviewers to state their opinions. This is done at aspect level of analysis using ontology. Then it determines whether they are positive or negative thereby giving a scaling system to identify the effectiveness of a product. So, we propose a technique here, Sentiment Analysis for Product rating through Ontology (SAPRO). This technique performs text mining techniques on the customer reviews obtained from any of these websites and calculates the rating for this product (out of 5), to establish the like-ability of the product by the existing customers. SAPRO considers the end users’ perspective while addressing the Sentiment Analysis problem. The research uses a combination approach of domain ontology and Stanford dependency relation which intends to enhance the sentiment classification. The proposed technique is programmed in Java 8 programming language taking review data from the Amazon website as the dataset which is in JSON format. The results show that the review rating calculated by the proposed technique has 91% similarity with the existing ratings from the customers. The difference is mainly because the customer gives discrete numeric ratings but explains his sentiment better in words. The results when compared to base research by Arindam Chaudharyet. al [24], clearly show that the proposed technique(91% accuracy) is better than base technique(80-88% accuracy) for reviews.