SK telecom has presented a new artificial intelligence (AI) recommendation technology at a leading global conference, highlighting its efforts to better understand and predict users’ true preferences.
The company announced on April 28 that its research paper on an AI recommendation model was accepted for presentation at ICLR 2026, one of the world’s most prestigious AI conferences alongside NeurIPS and ICML.
At the core of the research is a proprietary technology called Conflict-Aware Preference Optimization (C-APO). The model is designed to distinguish between conflicting user preferences within behavioral data and identify signals that more accurately reflect genuine user intent, enabling more precise recommendations.
The research team introduced a method that separates long-term, repeated user behaviors—defined as “consistent preferences”—from temporary or incidental actions, referred to as “superficial preferences.” When the two signals align, the model increases their importance; when they conflict, it treats the discrepancy as noise and reduces its influence. This approach allows the AI system to learn optimal recommendation criteria autonomously.
The new model also addresses a common limitation of traditional recommendation systems, which often misinterpret one-time user actions as long-term preferences. For example, if a user briefly engages with content outside their usual interests, existing systems may skew future recommendations. C-APO minimizes such distortions by prioritizing consistent behavioral patterns.
Another notable feature is improved explainability. Beyond simply recommending content or products, the system can provide logical reasoning behind its suggestions, a capability expected to enhance user trust in AI-driven services.
SK telecom plans to integrate the technology into its commercial services. The company aims to incorporate C-APO into an agent-based recommendation system that operates in three stages: interpreting user context, generating recommendation candidates, and validating and refining results. This structure is expected to significantly enhance personalization.
A company official said the research represents a foundational step toward understanding customers’ real preferences more accurately and added that the firm will continue to advance AI-driven customer experience innovation following its presentation at a global academic stage.
