What drives the helpfulness of online reviews? A deep learning study of sentiment analysis, pictorial content and reviewer expertise for mature destinations.

User-generated content (UGC) is a growing driver of destination choice. Drawing on dual-process theories on how individuals process information, this study focuses on the role of central and peripheral information processing routes in the formation of consumers’ perceptions of the helpfulness of online reviews. We carried out a two-step process to address the perceived helpfulness of user-generated content, a sentiment analysis using advanced machine-learning techniques (deep learning), and a regression analysis. We used a database of 2,023 comments posted on TripAdvisor about two iconic Venetian cultural attractions, St. Mark’s Square (an open, free attraction) and the Doge’s Palace (a museum which charges an entry fee). Following the application of deep-learning techniques, we first identified which factors influenced whether a review received a “helpful” vote by means of logistic regression. Second, we selected those reviews which received at least one helpful vote to identify, through linear regression, the significant determinants of TripAdvisor users’ voting behaviour. The results showed that reviewer expertise is an influential factor in both free and paid-for attractions, although the impact of central cues (sentiment polarity, subjectivity and pictorial content) is different in both attractions. Our study suggests that managers should look beyond individual ratings and focus on the sentiment analysis of online reviews, which are shown to be based on the nature of the attraction (free vs. paid-for).