The ways in which different components collectively influence review helpfulness is not well understood in hospitality. This study shows how multiple review data modalities can be integrated with big data analytics to develop an algorithm that predicts review helpfulness to benefit hospitality businesses. We specifically reconciled the joint effects of three components: metadata (influential factors), text descriptions (textual content), and images (user-provided photos). We then applied a conjoint deep learning framework to predict the number of helpfulness votes a review may receive. Our model was empirically validated using 138,177 reviews retrieved from Yelp at two time stamps; its predictive power exceeded that of models containing fewer or singular components. These findings underscore the importance of thoroughly capturing a review's informational value, including its text and visual content, to determine review helpfulness. We also highlighted the temporal effects of reviews by including a review's posting date in dynamic helpfulness measures. This automatic prediction model enables business managers to foresee helpfulness votes and detect helpful/unhelpful reviews in a timely manner.