Recurrent Neural Network for Map Query Suggestion Based on Hierarchical Context Attention

Abstract

Map query suggestion plays a critical role in improving user experience in location-based services by predicting users’ search intentions and recommending relevant queries. Traditional approaches often fail to effectively capture the complex temporal dependencies and multi-level contextual information embedded in user search behaviours. To address this limitation, this study proposes a Hierarchical Context Attention-based Recurrent Neural Network (HCA-RNN) within an encoder–decoder framework for map query suggestion. The proposed model learns sequential dependencies among queries within a session to capture short-term user intent, while simultaneously modelling cross-session relationships to represent long-term search context. A hierarchical attention mechanism is incorporated to selectively emphasize important queries and contextual information at different levels, enabling more accurate intent understanding. Experiments conducted on large-scale real-world map query logs demonstrate that the proposed approach significantly outperforms existing baseline methods in terms of standard evaluation metrics such as Recall and Mean Reciprocal Rank (MRR). The results confirm the effectiveness of hierarchical contextual attention in enhancing the accuracy and relevance of map query recommendations.