Evaluating Recommender System using Baseline Approaches
Rawat Sapna Jawansingh, Abhay Kumar · Procedia Computer Science · 2025
In the current era of rapidly advancing technology, the domain of recommendation systems built with sophisticated algorithms needs to be thoroughly assessed and evaluated using reliable baseline techniques. In this comprehensive work, a systematic evaluation of multiple baseline algorithms is carried out that helped to assess and categorize the methods based on the evaluation metrics and their performance like Content-Based Filtering (CBF) methods, Collaborative Filtering (CF), simple averaging methods, and hybrid methods. A Unified framework of performance metrics which includes Mean Absolute Error (MAE), F1-Score, Precision, Recall, Root Mean Squared Error (RMSE), and Normalized Discounted Cumulative Gain (NDCG) is used for measuring the performance of existing approaches, on the Movielens 1M, a publicly available stable dataset from GroupLens. Singular Value Decomposition (SVD) performed best with Root Mean Squared Error(RMSE) of 0.8757, while hybrid methods combining CBF and CF achieved an RMSE of 0.8745. Baseline methods, though simple, lacked personalization, highlighting the need for hybrid approaches. The results of this evaluation demonstrate significant limitations of averaging methods that lack personalization, content-based filtering methods require extensive feature engineering, collaborative filtering has scalability and cold-start concerns, and the existing hybrid methods however promising are complex and computationally costly. This study emphasizes the necessity of more advanced hybrid techniques that can improve personalization while simultaneously overcoming cold-start problems, providing more accurate, scalable solutions for real-world recommendation systems.