Recomendación de ítems a grupos de usuarios en base a la predicción de sus preferencias.

Abstract

Abstract: The present study provides a comprehensive analysis of recommender systems, exploring techniques such as collaborative filtering and algorithm hybridization. It focuses on the advanced prediction of group user preferences by employing machine learning algorithms and data analysis to identify patterns and similarities. The primary objective is to develop a model that generates accurate and relevant recommendations based on past behaviors, similar group characteristics, and predictions for new users, using aggregation methods that allow recommendations to be offered to a new user entering the system even if they were not considered during the initial clustering process, while ensuring that their information is recorded in the main dataset. Emphasis is placed on performance evaluation for prediction using metrics such as MAE and RMSE, and for recommendation using metrics such as precision, recall, and F1-score; the results of these metrics help determine the overall quality of the model, its predictions, and especially the recommendations it generates. Finally, with special attention to adapting recommendations for groups of users with similar characteristics, this work aims to improve the effectiveness of traditional systems by comparing the classical KNNBasic algorithm in its standard version with a variant that applies the same algorithm to different groups or clusters as appropriate.