Dealing with Ambiguity in Online Customer Reviews: The Topic-Sentiment Method for Automated Content Analysis

Abstract

Content analysis has become a widely used technique for the analysis of the large quantities of data that are generated online. Especially relevant for marketing researchers are customer reviews on websites such as TripAdvisor and Amazon, because they express customers’ satisfaction and they represent an important source of word-of-mouth for other consumers. Although the recent preference for sentence-constrained approaches has increased the accuracy of analytical methods, in many cases these methods still ignore some of the nuances contained within online reviews. In particular, current methods may not detect when a single topic is discussed both positively and negatively in a single review, or when a single sentence discusses two separate topics. The topic-sentiment method that is proposed in this paper addresses these two issues. It is a sentence-constrained approach that identifies ‘topic-sentiment pairs’; sentences that contain one word that describes the topic, and another that expresses the sentiment (positive or negative). To illustrate the analytical process, the method is applied to a dataset of 17,225 TripAdvisor reviews for restaurants in London. Results indicate that the topic-sentiment method offers a more nuanced approach for the analysis of customer reviews, while it retains the intuitiveness and simplicity of currently used methods.