Boosting or Attenuating? The Llinguistic Features of Sentiment Strength in User Generated Content

As markets have been labeled as conversations and consumer-to-consumer communications have been recognized as a dominant force in driving consumer patronage, companies are facing the challenge of having to resort to a new generation of performance metrics that provides guidance in a social commerce environment. With respect to this, sentiment analysis of text-based. User Generated Content (UCG) has become an increasingly popular way of staying in touch with customers. The viability of sentiment analysis as a performance metric, however, has been seriously questioned due to its limited predictive ability in assessing diverging degrees of positive and negative sentiments embedded in large and diverse volumes of online textual conversations. In this study, we advance automated text-mining modeling, based on linguistic theory, to deal with aforementioned issues. On the basis of emergent theorizing on speech acts, we zoom in on how linguistic style elements -modal and relational meaning- can boost or attenuate sentiment expression in online customer reviews. The modality of sentiment is accounted for by considering arousal intensity of affect-laden words. Moreover, commonalities of linguistic style elements on the basis of function words, such as pronouns, tenses and word interlinks (patterns) reflect conversants’ alliance or relational closeness to a consensual style of interaction. This is referred to as linguistic style matching. We test our theory-based predictions by examining more than 100.000 reviews across a range of 8 different product/service categories. The empirical results show that adding aforementioned linguistic elements increases the predictive ability of a sentiment classification model of customer online reviews, allowing firms to develop better quantitative metrics from online textual information. Finally, we corroborate our approach in sampled Facebook and Twitter conversations. Theoretical and managerial implications are discussed