Context Driven Bipolar Adjustment for Optimized Aspect Level Sentiment Analysis

Abstract

World Wide Web provides numerous opinionated data that can influence users. Reviews on online data highly affect the user’s perception while buying a particular or related product from an online shopping site. The online review provided by a customer helps other customers to make up their decision regarding purchasing that item. Looking at the developer’s and producer’s perspective, the opinions of customers on their manufactured items is helpful in identifying deformities as well as scope for improving its quality. Equipped with all this information, the product can be developed and managed more efficiently. Along with the overall rating of the product, the feature-based rating will have a great impact on the decision-making process of the customer. In this paper, an optimized scheme of aspect level sentiment analysis is presented to analyze the online reviews of a product. Reviews ratings have been used for learning approach. Inherently biased reviews are considered to optimize the Aspect Level Sentiment Analysis. Bi-polar aspect level sentiment analysis model has been trained using multiple kernels of support vector machine to optimize the results. Lexicon based aspect level sentiment analysis is performed first and later on the basis of bipolar words adjustment, and its effect on results, aspect level sentiment analysis for efficient optimization has been performed. A Web Crawler is developed to extract data from Amazon.
The results obtained outperformed traditional lexicon based Aspect Level Sentiment Analysis.