Knowledge Graph-based Conversational Recommender System in Travel

Conversational Recommender System (CRS) aims to accomplish recommendation tasks by interacting with users via natural languages. Different from the traditional recommendation, the communications with users in CRS enable the systems to capture users' latest preferences. Travel is one of the core application scenarios of CRS. However, current studies in travel mainly concentrate on restaurant recommendations, while the key aspect, tourism attraction remains unexplored due to the limitation of the dataset. Therefore, to explore CRS in tourism attraction and city recommendation, we contribute a high-quality dataset in the travel domain named CADT (City-Attraction Dataset for Travel), which consists of 3000+ tourism attractions and corresponding features with a hierarchical structure. We observe that the current state-of-the-art CRS method can hardly yield satisfactory performance on this dataset. To this end, we propose a simple but effective framework named RUI (Representation, Understanding, Interaction), which performs a path seeking process on the given knowledge graph to make travel recommendations. Both automatic and human evaluations are implemented, demonstrating the effectiveness of the RUI framework. Altogether, we introduce and explore the ways to build the first knowledge graph-based conversational recommender system focusing on city and tourism attraction recommendations in travel domain.