Effect of user-generated image on review helpfulness: Perspectives from object detection

Many e-commerce platforms encourage users to upload user-generated images when posting reviews, while it is still not clear what aspects in images are considered helpful by consumers. The purpose of this study is to empirically explore the effects of argument quality of images on review helpfulness, and how the effects are different across different product types. Inspired by the information adoption theory, this work measured the four important aspects (adequacy, accuracy, consistency, and relevance) for the quality of user-generated images by applying the object detection deep-learning methods. Based on the dataset from JD.com, our empirical results showed the effects of these aspects on review helpfulness across different product types. We found that accuracy and relevance of images positively affect review helpfulness on both product types. Surprisingly, consistency between images and text prompts review helpfulness for search products, while the effect is negative for experience products. Implications are also discussed.