Online reviews have received much attention in the study, among which the online review usefulness and how different characteristics of review content determine it are widely documented. However, current theoretical insights shed little light on the effects of visual features and how they correlate to review text. Thus, this study delves into the determinators of review usefulness, focusing on the textual and visual content as well as the interplay between them and the moderating role of commodity price.
Drawing on the Elaboration Likelihood Model (ELM), we develop a theoretical framework to examine how the complexity of textual and visual content in online reviews, the consistency between them and the visual quality of images affect their usefulness. We employ several deep learning techniques to extract textual and visual features from a dataset comprising 106,853 online reviews of 3,589 products on JD.hk.
Image-text consistency and visual quality of images (image aesthetics score and image clarity) positively influence review usefulness; image and text content complexity defined by the entropy of topic distribution and language model-based text sentence complexity, tend to have negative effects. Notably, for higher-priced products, visual attributes play a more significant role in shaping review usefulness than low-priced products, underscoring the importance of high-quality images in reviews of premium items.
This research enriches the existing literature on the review usefulness in e-commerce by elucidating the role of image and textual data through deep learning analyses, offering valuable insights for both theoretical understanding and practical application.
