An exploration of semantic tendencies in word of mouth business reviews

Explicit customer review ratings mark future business success. One important and well-studied aspect of customer satisfaction is a review's affective - positive or negative - valence. More recently, tools from natural language processing (NLP) applied to reviews show less obvious linguistic differences in review texts dependent on reviewer rating. Consistent with this is previous work using Linguistic Inquiry and Word Count (LIWC), showing that language use changes depending on one's current psychological state. Finer-grained analyses of review text focusing on less obvious linguistic categories may provide insight into customer values. In an attempt to explore how the content of a review is related to a review's explicit rating, we analyzed review texts using LIWC. LIWC determines the percentage of review text associated with a variety of different psychologically relevant categories such as social or cognitive words. We explore how certain categories of words relate to review ratings and use a support vector machine to determine how well each category predicts reviewer's review rating. We relate our findings to previous work and speculate that businesses would benefit from the application of various Natural Language Processing tools in attempting to obtain comprehensive insight into customer satisfaction. We end with the connection between this work and theories of language use, for which data sets of customer reviews may be useful for exploring the role of psychological state in determining word choice.