Scholarlink

The exponential growth of tourism-related information available through digital platforms has created challenges in filtering relevant data to provide personalized recommendations. This paper addresses the dynamic nature of user preferences in tourism by proposing two reinforcement learning (RL) models for dynamic user profiling: a Hierarchical Reinforcement Learning (HRL) model with Matrix Factorization and a Double Deep Q-Network (DDQN) model. These models aim to enhance the recommendation of Points of Interest (POIs) by adapting user profiles based on evolving interests across Topics of Interest (TOIs). Experimental evaluations, using an adapted version of MovieLens 100K dataset, demonstrate that the HRL model excels in updating profiles for users with extensive interaction history, while the DDQN model is more effective for new or less active users. This work highlights the potential of RL models for dynamic user profiling in improving personalized tourism recommendations and offers insights into the suitability of different models based on user interaction patterns.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic User Profiling for Personalized Tourism Recommendations Using Reinforcement Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tommaso Ferrario,
  • Elisabetta Fersini,
  • Enza Messina,
  • Gabriele Sormani

摘要

The exponential growth of tourism-related information available through digital platforms has created challenges in filtering relevant data to provide personalized recommendations. This paper addresses the dynamic nature of user preferences in tourism by proposing two reinforcement learning (RL) models for dynamic user profiling: a Hierarchical Reinforcement Learning (HRL) model with Matrix Factorization and a Double Deep Q-Network (DDQN) model. These models aim to enhance the recommendation of Points of Interest (POIs) by adapting user profiles based on evolving interests across Topics of Interest (TOIs). Experimental evaluations, using an adapted version of MovieLens 100K dataset, demonstrate that the HRL model excels in updating profiles for users with extensive interaction history, while the DDQN model is more effective for new or less active users. This work highlights the potential of RL models for dynamic user profiling in improving personalized tourism recommendations and offers insights into the suitability of different models based on user interaction patterns.