Evaluating CVIM Performance in Convenience Goods Through Transformer-Based Text Analytics

Abstract—The emergence of e-commerce sites has also facilitated the consumers in giving an easy review of the products they bought. Though the amount of consumer reviews is large and their number is constantly increasing, the data is not directly applicable as the basis of decision-making. The reason is that reviews are not structured. This study use BERT for sentiment analysis and topic modelling. Then develop Customer Voice Impact Matrix (CVIM) to illustrates the effect of the re- view topics on the perception of the customers by charting the average sentiment in the horizontal axis and frequency of the topic in the vertical axis. Result show that CVIM was succesfully developed and give valuable insight for decision making in businesses. By using CVIM, consumer feedback that was originally unstructured becomes easy to analyze by mapping topics into 4 quadrants, namely Fix Urgently, Promote Strength, Hidden Gem and Minor Issue. Index Terms—Natural Language Processing (NLP), Sentiment Analysis, Topic Modeling, Customer Voice Impact Matrix (CVIM), BERT, BERTopic.