Exploring the occurrence of a point of diminishing returns in sentiment analysis of customer reviews

Abstract

Customer reviews have been proven to have an influence on purchase decisions made by consumer in a market where online shopping is becoming more and more common and web merchants become strong competitors for traditional sales channels. Since the number of stars a product receives (rating) has been found in research not to be the only influential factor for purchase decisions, a line of research has been being applied over the recent years to automate the analysis of the review texts and interpret the sentiment expressed in those texts via sentiment analysis. The number of customer reviews has been growing with the increase of online sales so companies deciding to use sentiment analysis face an increasing number of reviews to be analysed. With the challenges that an increased number of reviews can present for companies and researchers a question surged: Does the average sentiment polarity towards a product stabilize after a number of customer reviews, making the analysis of additional reviews redundant? The goal of this thesis is to explore this question in an effort to understand if analysing the total number of reviews of a product is really necessary to measure the accumulated sentiment polarity or if measuring a smaller number might produce the same results. The results show that there is an association between the number of customer reviews and the appearance of a point of diminishing returns for the accumulated sentiment polarity, supporting the notion that a company trying to rate the sentiment polarity towards a product could decide not to evaluate all the customer reviews but a fraction of them or stop processing them at a point, optimizing the resources dedicated to sentiment analysis.