Exploring customers’ perceptions toward green restaurants using user-generated content

Abstract

Many restaurants have incorporated sustainable or “green” practices to minimize environmental harm and to build a positive brand image. However, customers may not always appraise such efforts but rely on their existing images to implicitly process what they perceive, resulting in differences between the images that a company seeks to communicate and what customers actually perceive. Therefore, the purpose of this study was to identify green images from the free-recalled user-generated content (UGC) of certified green restaurant customers. First, the salient image categories were extracted, and effects of reviewer and restaurant characteristics on the recalled green image were examined. Then, the image network structures including both higher- and lower-level image elements and their characteristics were investigated. Post-visit online reviews (N=25,098) of 70 certified green restaurants, written between March 2014 to February 2019, were selected from TripAdvisor.com to capture the free-recalled green restaurant images expressed in unstructured texts. After typical data preprocessing, 51 salient image categories were identified using the structural topic model (STM) algorithm followed by a factorial MANCOVA and LSD post hoc analysis to estimate the effects of reviewer and restaurant characteristics on the green image. A topic-level network was drawn based on topic proportion correlation matrix, and a green image network structure was examined based on the co-occurrence of the unique words found in the UGC. For both networks, a community detection algorithm was applied to discover the subgroups from the image associations. In addition, the image nodes were classified into three groups (i.e., core, semi-periphery, or periphery) based on eigenvector scores, and sentiment and emotion scores were assessed for each image node. Both general restaurant attributes (e.g., food, service, atmosphere, and value) and green attributes emerged from the STM. Some specific restaurant attributes (e.g., employees’ attire) used in previous studies did not emerge as a relevant topic. The extent of green practice implementation (p<.001) and the duration of the certification program (p<.001) were significantly associated with the likelihood of the customers mentioning a green practice topic, and female customers mentioned more about sustainable foods than males (p<.001). In the topic-level network, positive image categories (e.g., T44, satisfaction and T5, good flavor) tended to have higher eigenvector scores (>0.99) than negative categories (e.g., T6, T46 related to bad service; eigenvector scores<0.22), indicating that positive topics were more easily recalled among customers. Similarly, the green image network and image associations relevant to green attributes contained positive sentiment scores (>0.95). While the majority of food-focused green image associations were classified as core or semi-periphery, environment-focused green image was classified as periphery. The results demonstrated that the food-focused green image associations were more tightly connected to other image associations and more likely to be activated than environment-focused images. This study tested the category-based perspective and associative network model with the free-recalled UGC to conceptualize the green restaurant image. Various machine-learning based approaches and network analysis improved reproducibility and overcame subjectivity in traditional qualitative analysis. Based on the findings, restaurateurs may develop green marketing strategies to gain competitive advantages.