This paper presents a multimodal deep learning–based tourism recommendation system designed to address the challenges of querying information related to administrative boundary changes following local government mergers in Vietnam. The main contributions of this study include: (i) constructing a standardized administrative data repository derived from national legal documents; (ii) applying deep learning models for Vietnamese text processing and speech recognition to analyze user queries from both text and voice inputs; and (iii) integrating a Knowledge Graph to model and query complex relationships between old and new geographical entities. Furthermore, we developed a demo application that enables users to search for and receive tourism recommendations through multimodal queries. Experimental results on the administrative dataset demonstrate that the proposed system can provide relevant information on demand and deliver contextually appropriate search results within the Vietnamese setting.