With the continuous proliferation of deepfake information in AI-generated content, it is imperative to accurately identify and assess users' deficiencies in coping with such risks, which provides a critical foundation for enhancing users' capabilities in managing deepfake information risks. Drawing on risk management theory, this paper integrates the three fundamental stages-risk identification, evaluation, and response-to construct a comprehensive indicator system for assessing deepfake information risk management capability. A risk assessment model is developed based on the five-element connection number framework of similarity, difference, and opposition, combining the Uncertainty Hierarchy Method with Set Pair Analysis. The model's effectiveness is demonstrated and validated through empirical case analysis. The empirical findings indicate that users' overall risk management capabilities regarding deepfake information remain at an average level, with only a weak tendency toward further decline, suggesting a relatively low risk of capability deterioration. Specifically, users' capabilities in risk identification, evaluation, and response are all situated within the anti-trend zone of negative development. Among these, the downward trend in risk response capability is the most pronounced, underscoring the need for targeted measures to optimize this dimension.