This research explores counterfactual explanations in Sequential Recommender Systems (SRSs), investigating how minimal perturbations in user interaction history influence recommendation outcomes.

Methodology

• 01 Genetic Algorithms (GA): Custom-optimized for discrete item sequences to explore alternative user interaction paths effectively.
• 02 Automata Learning (AL): Used to construct interpretable surrogate models that provide transparency into complex decision-making processes.

Results & Evaluation

The framework was evaluated on the **MovieLens 100K** and **1M** datasets, confirming high fidelity in maintaining original recommendation accuracy while providing actionable counterfactual insights.