AN INTELLIGENT SENTIMENT ANALYSIS SYSTEM FOR MEASURING CUSTOMER LOYALTY AND MAKING DECISIONS BASED ON FUZZY LOGIC

Abstract

This paper presents an intelligent approach of measuring customer loyalty to a specific product based on the analysis of comments. General sentiment analysis in tweets and messages is quite common, but task-oriented analysis of user opinions and measuring their level of loyalty is a new idea. The tricky part of doing task-oriented sentiment analysis lies in measuring customer loyalty to a particular product based on how customers feel about that product itself. The resulting data on the level of customer loyalty to the product can help a new customer to make a decision on a specific product, taking into account its various characteristics and feedback from previous customers. The dataset was a large dataset of online customer testimonials from Amazon.com. The set of initial data is a set of reviews, from which the proposed approach forms an aggregated assessment of opinions, then a fuzzy logic model is used to measure customer loyalty to the product. In the proposed approach, the input text is first processed using such methods as tokenization, removal of stop words, lemmatization, then the parts of speech are marked and the polarity of the reviews is analyzed, then fuzzy logic methods are applied to the obtained aggregated estimates to determine the degree of customer loyalty to the product. This work used various open API libraries such as SentiWordNet, Stanford CoreNLP, etc. The approach used focuses on identifying the sentiment of reviews, which can be positive, negative and neutral. In our study, we used a triangular membership function, also known as trimf, because it supports three variables and creates a relationship between them. The implementation of the approach ensures high accuracy in determining loyalty to e-commerce products, which is superior to previous approaches, and the use of fuzzy logic has significantly increased the values of such indicators as precision, recall, and F-measure.