Dialogue-Based User Needs Extraction for Effective Service Personalization

The research of service personalization is flourishing due to the development of machine learning and natural language processing. Despite the prevalence of prior research based on deep learning and dialogue, it remains challenging to reconcile the disadvantages of machine learning, such as explainability, with the strength of utilizing big data. This research proposes a user needs model that incorporates three elements: user readability, ease of extraction through dialogue, and the potential for advancement in machine learning. Additionally, a voice dialogue-based extraction method is designed and constructed to extract the proposed needs. Specifically, by adopting the 6W1H format for the needs model, a simple yet powerful dialogue flow is achieved and enables a comparison of existing services and needs simultaneously. The main modules of the system are a voice dialogue agent, a dialogue system, and a natural language processing-based needs extraction API. Through designing, implementing, and integrating each module, this study realizes a needs extraction system in Japanese. Furthermore, by operating the realized system, a simple evaluation of the needs model and the system is carried out. As a result of this research, both the user and the system can extract needs that are highly readable, contributing to the realization of user-friendly and effective service personalization.