Predicting User Interest Using Hierarchical-Based Clustering for Recommender Systems

In the realm of personalized recommendation systems, accurately predicting a user's interest in specific items is paramount. In this study, we aim to develop an estimation method that predicts such interests with sufficient accuracy. Drawing inspiration from human decision-making processes, our method considers two primary sources of information: a user's voting behavior and their similarity to other users. By initializing all unrated items based on average user ratings and employing hierarchical clustering to group users, we can anticipate a user's rating for uncommented items. We evaluated our approach using the MovieLens database, which, due to its vast number of unrated items, presents a significant challenge. The results confirm the efficacy of our method in accurately estimating user interests due to RMSE-percent = 51.012.