Demystifying Sequential Recommendations: Counterfactual Explanations via Genetic Algorithms

Sequential recommendation systems are "black boxes" that need better explanations: These AI systems effectively predict what users want next, but their complexity makes it difficult to understand why they make specific recommendations—this paper addresses that gap by showing users what changes to their past behavior would alter recommendations.

The paper introduces a practical "what-if" explanation method: Rather than abstract technical explanations, the approach answers concrete questions like "What if I had watched different movies last week?"—helping users understand the logic behind recommendations in intuitive terms.

Generating these explanations is computationally complex: The researchers prove that creating counterfactual explanations for sequential data is an NP-Complete problem (meaning it's extremely difficult to solve), yet they still developed a genetic algorithm that successfully generates meaningful explanations while maintaining accuracy close to 100%.

The method was rigorously tested across multiple scenarios: The team evaluated their approach using three different datasets and three different recommendation models across four experimental settings, demonstrating that the explanations remain interpretable and trustworthy across varied real-world conditions.

This work advances transparency and user trust in AI systems: By enabling users to understand why they received specific recommendations through concrete "what-if" scenarios, the research contributes to making AI systems more accountable and trustworthy—key goals for responsible AI deployment.