Multi-level Sentiment Information Extraction Using the CRbSA Algorithm

Abstract

Social network platforms allow the customers to feedback and complain about their opinions on products and services. Normally, users' feedbacks on social networks are unstructured data usually involving an enormous size of texts, called Social Big Data. Even though Social Big Data supports marketers by giving the information about the customers' sentiments, a lot of organizations suffer with labor intensive and time-consuming tasks in extracting the customers' satisfaction from Social Big Data manually. Therefore, an automatic process to extract the information from Social Big Data is required by marketers and decision-makers. To deal with this requirement, this paper proposes a new sentiment information extraction algorithm, called the Contrast Rule-based Sentiment Analysis algorithm that intends to extract the information automatically. We prove the validity of our proposed algorithm through comparison with the well-known sentiment information extraction algorithms, general word counting and SentiStrength. Applying on the labelled customer feedbacks on the Amazon dataset, our algorithm extracted sentiments more correctly than the general word counting and SentiStrength algorithms, especially in the negative cases. The processing time is also faster than the SentiStrength algorithm. This algorithm can be applied in a marketing system to help extract the customers' satisfaction, especially work as an alarming tool for negative comments.