Abstract
In the marketing field, online service providers have recently implemented personalized recommendations on various mobile applications as a part of customer relationship management. However, there has long been an issue of consumer heterogeneity, where each customer has internal differences that are difficult to discern from behavioral logs. On the other hand, the transfer of pre-trained model referred as large-scale language models (LLMs) has facilitated text data analysis, such as customers' reviews on the online platform, wherein they express their reasons for the evaluations which cannot be obtained from behavior logs. Therefore, in this study, we introduce a conceptual model of multimodal deep learning, combining review texts with traditional customer and store information. To rephrase, we develop a restaurant evaluation model that integrates text data analysis to comprehend consumer heterogeneity, alongside conventional analytical methods. Our comprehensive exploration and comparison of multiple models reveal that the proposed model shows the best prediction accuracy. Moreover, we discuss the limitations of analyses relying solely on traditional customer information and the potential for advancements in future purchase prediction models.
