A Hybrid GA–TLBO Optimization Framework for Collaborative Filtering with a Novel Gravitational Similarity Metric

Overview

Collaborative filtering recommends items by finding users with similar tastes. This paper improves both halves of that process: how similarity between users is measured, and how the model’s weights are optimized.

Method

The framework combines a genetic algorithm (GA) with teaching–learning-based optimization (TLBO) into a single hybrid optimizer, and introduces a new gravitational similarity metric that models the “pull” between users based on their rating behavior. I contributed substantially to the implementation of both the hybrid optimizer and the similarity metric.

Results

The hybrid approach reduces prediction error by up to 11.6% compared with GA and TLBO baselines, and achieves the lowest mean absolute error (MAE) under 10-fold cross-validation.

Experiments

I designed and executed the large-scale evaluations behind these results, totaling more than 972 compute hours.