Transformer-Based Multi-task Learning for Queuing Time Aware Next POI Recommendation

Next point-of-interest (POI) recommendation is an important and challenging problem due to different contextual information and wide variety in human mobility patterns. Most of the prior studies incorporated user travel spatiotemporal andsequential patterns to recommend next POIs. However, few of these previous approaches considered the queuing time at POIs and its influence on user's mobility. The queuing time plays a significant role in affecting user mobility behaviour, e.g., having to queue a long time to enter a POI might reduce visitor's enjoyment. Recently, attention based recurrent neural networks-based approaches show promising performance in next POI recommendation but they are limited to single head attention which can have difficulty finding the appropriate complex connections between users, previous travel history and POI information. In this research, we present a problem of queuing time aware next POI recommendation and demonstrate how it is non-trivial to both recommend a next POI and simultaneously predict its queuing time. To solve this problem, we propose a multi-task, multi head attention transformer model called TLR-M. The model recommends next POIs to the target users and predicts queuing time to access the POIs simultaneously. By utilizing multi-head attention, the TLR-M model can integrate long range dependencies between any two POI visit efficiently and evaluate their contribution to select next POIs and to predict queuing time. Extensive experiments on eight real datasets show that the proposed model outperforms than the state-of-the-art baseline approaches in terms of precision, recall and F1 score evaluation metrics. The model also predicts and minimizes the queuing time effectively.

Automated summary

The study introduces a novel problem of recommending next POIs while predicting their queuing times simultaneously. The TLR-M model outperforms existing methods in efficiently integrating long-range dependencies between POI visits, selecting next POIs, and predicting queuing times. The extensive experiments on real datasets demonstrate the model's effectiveness in terms of precision, recall, F1 score metrics, and minimizing queuing times.