The effects of the aesthetics and composition of hotels' digital photo images on online booking decisions

The study explores the influence of visual elements in hotel digital photography on online booking decisions. Neuromarketing research highlights the brain's rapid processing of visual information, making it a crucial aspect of marketing. While prior research acknowledges the importance of hotel images, methodological limitations in using surveys and experiments have hindered a deep understanding of the specific aesthetic and compositional elements driving booking behavior. This study addresses these limitations by employing AI-powered computer vision techniques to analyze a large dataset of hotel images, aiming to identify specific image properties that correlate with higher booking rates. The central research question is: What aesthetics, composition, and image elements relevant to hotel photography determine positive effects on online hotel bookings?

The literature review examines computational aesthetics, photography elements and principles, and the role of digital images in hotel e-commerce. Computational aesthetics aims to predict human responses to visual art, while photographic composition focuses on arranging elements within a frame to capture attention. In hotel marketing, high-quality images are crucial for conveying intangible aspects of the hotel experience, reducing cognitive load, and building customer confidence. However, excessive images can lead to 'choice paralysis,' while misleading images can negatively affect bookings. Existing research has highlighted the positive effects of larger images, the impact of human presence, and the importance of image positioning on websites, but lacks a comprehensive methodology for analyzing aesthetics and composition's influence on actual bookings.

Data was collected using a random multi-stage sampling method, selecting 225 4-star hotels in Barcelona and gathering 6 images per hotel (totaling 1350 images) from Booking.com. The binary target variable ('booked' or 'not booked' within the last hour) was obtained for each hotel. Google's Inception v3 neural network performed image embedding, generating 2047 image descriptors. Logistic regression, support vector machines (SVM), and a multilayer perceptron (MLP) neural network were employed to build predictive models of booking based on image descriptors. Fuzzy cognitive mapping was then used to analyze the 'selling' properties of high-probability booking images (those with a probability ≥0.6 according to logistic regression), examining factors like time of day, angle, human presence, and color scheme.

Google's Inception v3 model generated 2047 image descriptors, which served as feature variables in the predictive models. The MLP neural network outperformed logistic regression and SVM in terms of AUC, CA, F1 score, precision, and recall (Table 2, Fig. 2). The MLP model achieved an accuracy of over 83%, indicating reasonable predictive power. Analysis of the confusion matrix (Fig. 3) provided insights into misclassifications. Fuzzy cognitive mapping (Table 3, Fig. 5) analyzed the 'selling' properties of images with high booking probabilities (≥0.6). Key findings from this analysis include: For exterior shots, no human presence and a monochromatic color scheme are beneficial; for lobby shots, the presence of staff is important; and for room shots, natural light, angled shots from a corner, and a subtle color scheme were positively associated with booking probabilities. The fuzzy cognitive map visually represents the relationships between these factors.

The study's findings advance marketing theory by applying AI to analyze the complex interplay between hotel image aesthetics and booking behavior. Unlike prior research relying solely on surveys and experiments, this study's AI-powered approach offers a more objective and nuanced understanding of the visual elements that influence bookings. The results confirm the importance of previously identified factors while providing more specific insights into optimal image composition and aesthetics. The developed model can predict booking probabilities based on image features, showcasing the potential of computer vision for analyzing online visual content in marketing. The study highlights the role of hotel photos as a supplementary factor influencing booking decisions alongside pricing, location, and reviews.

This research demonstrates the effectiveness of AI-powered computer vision in identifying specific image properties that predict hotel bookings. The findings provide practical recommendations for hotel marketers to create more effective online visual content. Future research could involve larger datasets, specialized neural networks trained on hotel images, inclusion of control variables (e.g., hotel star rating, price), and exploration of other destinations to improve generalizability. Advanced techniques like Generative Adversarial Networks (GANs) could also enhance future studies.

The study's 83% model accuracy could be improved with larger datasets and a specialized neural network trained on hotel images. The focus on Barcelona limits generalizability. The inclusion of control variables and the use of more advanced computer vision techniques would strengthen future studies. The reliance on fuzzy cognitive mapping, which is subjective, is another limitation.