Hybrid Recommendation System Based on Implicit Feedback with Collaborative Filtering and Gradient Boosting

Abstract

Recommendation systems are essential components in video streaming services as they assist users in selecting relevant content in line with the increasing availability of large-scale content. However, most recommendation systems still rely on explicit feedback data such as ratings, which are often unavailable on many platforms. This study aims to develop a hybrid recommendation system based on implicit feedback by constructing an interaction score derived from user behavior as a substitute for ratings. The proposed model integrates collaborative filtering methods (matrix factorization and k-nearest neighbor) with the CatBoost gradient boosting decision tree algorithm. The evaluation was conducted using empirical data from a video streaming service, with performance measured using root mean squared error (RMSE) and mean absolute error (MAE). The results indicate that the hybrid model achieves lower RMSE and MAE values compared to individual models. These findings confirm that the hybrid approach is effective in improving recommendation accuracy while also contributing to enhanced user experience quality in video streaming platforms without explicit rating data.