Abstract:Online reviews and customer ratings are used by hotel booking service providers to assist customers in finding a hotel of interest as rapidly as they can. Reviews, however, offer a more thorough understanding of the hotel, but most visitors' lack of time or patience to read them all. Hence, in this era, there are various hotel recommendation systems which support the customer by recommending the hotels based on their scenarios. Gated recurrent units (GRU), soft Q-function, and maximum entropy reinforcement learning (RL) are integrated in this study to form a PCA clustering based hotel recommendation model. Gated recurrent units (GRU) have been established for gathering user preference data based on their initial query and its reply. Using the state representation as input, the actor would provide the most effective action for the present state. Two different sorts of activities are being done by the actor, such as providing a recommended list of items and asking a request question. With stochastic gradient descent-based alternation between optimising networks, the function approximators will be employed in this study for the soft Q-function and the policy to undertake a balanced search between exploration and exploitation. Its goal is to maximise both the policy entropy and the anticipated future return in order to increase the exploration efficiency of the policy by maximum entropy reinforcement learning. The temperature parameter determines how important the entropy term versus the reward. The experimental finding illustrates that the suggested algorithm works superior than the conventional collaborative filtering method using both error and prediction measures.

Enhanced entropy based reinforcement learning hotel recommendation system