Parallelized Movie Recommendation System for Efficient User Personalization
Journal
2025 33rd Telecommunications Forum (TELFOR)
Date Issued
2025-11-25
Author(s)
Petrushevski, Darko
DOI
10.1109/telfor67910.2025.11314437
Abstract
In today's data-rich environment, recommendation systems must process vast user-item interactions efficiently. We present a parallel movie recommendation pipeline using Apache Spark's MLlib, accelerating training and prediction with distributed scheduling. Partitioning the rating matrix yields nearlinear speedups. On the MovieLens 32 M dataset [1], our ALS model runs 99% faster than a sequential baseline on 10 cores, with an RMSE of 0.82. We also analyze performance factors such as cores, shuffle, block size and caching.