Work

Leveraging Clustering and Reinforcement Learning for a Product Recommender System

Year: 2026
Type: article
Abstract: The rapid development of digital platforms and e-commerce necessitates the adoption of strong recommender systems that can accommodate a wide range of user preferences. Traditional recommender systems often rely on collaborative filtering and content-based methods, which cannot grasp the complex relationships between users and products. Systems struggle with issues including sparse data, little flexibility, and inadequate user customization. The static nature of collaborative filtering and content-based filtering restricts their efficacy in dynamic situations, where item availability and user preferences can change. Here we are proposing a reinforcement-learning-based method that operates on the clustered data, where the agent learns optimal strategies to recommend products by navigating a defined action space. The proposed method employs a grid-based environment where actions determine the agent's progression through clusters. A variety of clustering algorithms were applied, and the one with the minimum standard deviation was selected for the formulation of states of a grid-based environment. Rewards are assigned based on user satisfaction with the recommended items, encouraging the agent to make relevant and impactful recommendations. Evaluation is done on two public datasets, Amazon and Movie Lens. The evaluation parameters used were start state count analysis, number of states visited by each user, return earned to reach a goal state from a predetermined start state, and number of steps taken in each episode during the learning process.
Language: English
Open Access status: diamond
OpenAlex ID: W7202201209
FWCI:
Cites: 33
Cited by:
Related to: 10