Dynamic POI Recommendation for Time-Efficient Urban Tourism Using Deep Reinforcement Learning

 Suggested by: Omid Reza Abbasi, Johannes Scholz

Keywords: Deep Reinforcement Learning, Real-Time, Agents

Objective:

1. Develop a DRL framework that models tourist state as a function of location, time, weather, and past interactions. 

2. Design a reward function that balances user satisfaction with time efficiency. 

3. Integrate real-time data sources (e.g., weather APIs, POI databases, OpenStreetMap) to inform the DRL agent's decision-making. 

4. Evaluate the system's performance through simulations and user studies, focusing on time efficiency and user satisfaction. 

Short Description: Urban tourism often presents challenges in maximizing the visitor's experience within limited time constraints. This is particularly true in culturally rich cities, where a plethora of historical sites, musical events, and scenic attractions exist for the tourist's attention. Traditional recommendation systems often fall short in adapting to the dynamic nature of tourist preferences, influenced by factors such as time of day, weather conditions, and real-time events. This thesis proposes an approach utilizing Deep Reinforcement Learning (DRL) to create a dynamic Point of Interest (POI) recommendation model. The goal is to optimize tourist itineraries for time efficiency, ensuring visitors experience the best of a city while adapting to real-time changes. Existing research in tourism recommendation systems often relies on collaborative filtering or content-based methods, which struggle to capture the dynamic nature of tourist preferences. Recent advancements in DRL offer a promising approach for creating adaptive and personalized recommendation systems. This thesis will explore the application of DRL techniques, such as DQN or actor-critic methods, to the specific context of urban tourism. It will also examine the integration of contextual data sources and the design of effective reward functions for time-sensitive recommendations. 

Start: Anytime

Relevant Studies:

Afsar, M. Mehdi, Trafford Crump, and Behrouz Far. "Reinforcement learning based recommender systems: A survey." ACM Computing Surveys 55.7 (2022): 1-38. 

Chen, Lei, et al. "Multi-objective reinforcement learning approach for trip recommendation." Expert Systems with Applications 226 (2023): 120145.