Customer Feedback Analysis Applying the RFC Model Hybrid Sentiment Extraction and Contrast Dictionary

Abstract

In recent years, the marketers have been moving to sale and trade their products in social markets due to the growth of Internet technologies. This allows the consumers to generate, share, criticize, and feedback on their interested products. The customer feedbacks come from many sources and the data become massive. The so- called Social Market Big Data (SMBD) involves many types of customers' expressions which are valuable to extract their opinions and satisfactions to support the marketers to make better decisions. However, the characteristics of SMBD are represented in unstructured formats which usually are only human readable. Manual information extraction is time consuming and labor-intensive. In this study, a new algorithm multi- level sentiment analysis to extract the customer opinions from feedbacks, called contrast dictionary is proposed. It can perform better than two well-known algorithms, the SentiStrength and Word-count, especially on negative feedbacks. Furthermore, a new hybrid approach to extract sentiments which improves the performance of sentiment extraction algorithms, called the aspect-based sentiment information extraction is also proposed. Moreover, the marketers require more comprehensive information on customer perspectives concerning the products and services comparing with the others. Previously, there had been no studies to grade the products based on the comparison of extracted information from SMBD. In relation to that, the RFM (Recency, Frequency, and Monetary) model is a measurement technique to compare market information, especially in traditional market analytics. This research also proposes a new approach by modifying the RFM model to classify the products from SMBD, called the RFC (Recency, Frequency, and Credit) model. The model focuses on the social market information and product categorization based on the customer satisfactions from customer feedbacks. The performance of the contrast dictionary has been validated with well- known sentiment information extraction algorithms, SentiStrength and Word-count, tested on Yelp and Amazon review polarity datasets. The proposed algorithm yields the 76.09% accuracy on the Yelp dataset, comparing with the 73.79% accuracy of SentiStrength and the 69.38% accuracy of Word-count. Also, it yields the 72.58% accuracy on the Amazon dataset that is more correct than the SentiStrength and Word- count those yield the 69.68% and 67.35% accuracies respectively. Furthermore, the proposed hybrid sentiment approach can improve the accuracy about 6.12% on the training dataset and 11.67% on the testing dataset. The RFC model produces new knowledges from customer feedbacks on products to be applied on a decision support and recommendation system in marketing management.