Abstract

In this study, we propose a novel model to design dynamic hotel room pricing strategies that consider the specific requirements associated with the tourism sector. Reinforcement learning (RL) is used to formulate the problem as a Markov decision process (MDP) and Q-learning is used to solve this problem with a new reward function for hotel room pricing which considers both the profit and demand. In the proposed model, the basic features of the hotels are digitized and expressed in a way that similar hotels get close values. In this way, price predictions for the hotels that are newly included in the system can be made through similar hotels and the cold start problem is solved. In order to observe the performance of the proposed model, we used a real-world dataset provided by a tourism agency in Turkey and the results show that the proposed model achieves less mean absolute percentage error on test data. In addition, we also observe the training phase and show that the proposed RL method has smooth reward transitions between timesteps and has a reward curve more similar to the desired exponential rise compared to recently recommended RL models with different reward functions in dynamic pricing.

摘要

在本研究中，我们提出了一种新颖的模型来设计动态酒店客房定价策略，该策略考虑了与旅游业相关的具体要求。使用强化学习 (RL) 将问题表述为马尔可夫决策过程 (MDP)，并使用 Q 学习通过考虑利润和需求的酒店客房定价新奖励函数来解决此问题。在所提出的模型中，酒店的基本特征被数字化并以类似酒店获得接近值的方式表达。这样就可以通过同类酒店对新纳入系统的酒店进行价格预测，解决了冷启动问题。为了观察所提出模型的性能，我们使用了土耳其一家旅游机构提供的真实世界数据集，结果表明，所提出的模型在测试数据上实现了较小的平均绝对百分比误差。此外，我们还观察了训练阶段，结果表明，与最近推荐的动态定价中具有不同奖励函数的 RL 模型相比，所提出的 RL 方法在时间步之间具有平滑的奖励过渡，并且奖励曲线更类似于所需的指数上升。