LAUSR

Abstract Spatial keyword query is an important technique for recommending users their desired POIs in self-driving services. Inferring query intention has been recognized as an important yet challenging issue for spatial keyword search. However, existing methods are inadequate to discover qualified results due to the inability to capture the intention of short-text input keywords. In this paper, we adopt a conceptual inference based method to deduce implicit intentions of users and thus are able to find more meaningful answers. Firstly, a locality-aware inference model is designed to generate concepts by considering typicality, granularity and spatial distribution, taking into account the hypernym–hyponym relationships in knowledge graphs. Afterwards, we propose a novel interactive framework to learn conceptual preferences for users, by using k-skyband to prune unpromising objects at the beginning and employing dense subgraph to select promising candidates in each interaction round. After a small number of rounds of learning, all objects can be rationally ordered by a user’s personalized ranking function which is unknown in advance. Empirical study on two real datasets demonstrates the effectiveness of our proposed conceptual inference and preference learning based methods.