Deep Reinforcement Learning-Based Recommendation System for Libraries Using Implicit Feedback

Abstract

The vast collections within library networks can make it difficult for users to find books that match their interests. This project addresses this challenge by developing a prototype recommendation system (RS) for the library network of the Diputació de Barcelona, powered by Deep Reinforcement Learning (DRL). The system personalises book recommendations based on users' borrowing history and profiles, overcoming the lack of explicit ratings through implicit feedback mechanisms. The study addresses two key challenges commonly faced by recommender systems: providing personalised suggestions that match individual preferences, and ensuring diversity in recommendations to avoid repetition or overly popular content. The recommendation task is framed as a Markov Decision Process (MDP), selected after an extensive review of state-of-the-art reinforcement learning-based recommender systems (RL-based RS). By applying DRL, the system aims to optimise long-term user engagement by focusing on cumulative rewards rather than immediate results. The project involves extensive data analysis, DRL model design, and iterative experimentation to refine the system's performance. This work helps to advance the application of DRL in the context of library systems and demonstrates its potential to enhance personalised user experiences in recommendation tasks. Although the project highlights the potential of DRL in library RS, challenges such as the high-dimensional action space and the scarcity of interaction data from real library environments prevented the model from achieving its initial performance goals. These limitations emphasise the inherent complexity of implementing DRL in large-scale, sparsely populated environments.