Understanding Consumer Behavior in Electronic Commerce with Image Sentiment

Customers in electronic commerce translate available information into buying decisions based on needs and behavioural tendencies. For this purpose, shopping sites provide detailed information on products and services ranging from simple figures (e.g., dimensions) to textual descriptions, user-generated reviews and visual images. All of the aforementioned sources of information may have a substantial impact on the decision-making of customers.However, little is known about the ways in which visual information is processed and used in decision-making. As a remedy, this research seeks to shed light on the informativeness of visual content in electronic commerce. This objective represents a highly relevant area of research, since visual information plays a major role in determining price, choice and thus willingness to buy (see [1], [2]). Unfortunately, such an undertaking is challenging as we are not aware of any existing methods for measuring the degree of informativeness of images. The main focus of visual sentiment analysis has been on the classification of the sentiment of images stemming from a fairly heterogeneous set. In contrast, we aim to build a model addressing a somewhat homogeneous set of images. In our case, we use photographs from real estate listings in order to predict the rent price. As an immediate implication, our findings show operators of e-commerce or recommender systems how to optimize the presentation of their products and services.