<p>The rapid growth of mobile data traffic has exacerbated challenges for cellular networks, including backhaul congestion and limited base station (BS) capacity. Wireless edge caching, which involves storing content closer to users at the network edge, presents a promising solution. In this context, device-to-device (D2D) communication plays a key role by enabling mobile devices to exchange data directly, reducing reliance on the cellular infrastructure. This paper addresses the uncertainties in D2D-based content caching, particularly content popularity fluctuations and channel variability. We propose a robust combinatorial optimization framework for content caching in relay-assisted D2D networks, optimizing both content placement and relay selection under uncertainty. Our approach models content placement as a 0–1 integer programming problem, leveraging weighted b-matching to derive optimal caching strategies under nominal conditions. Additionally, for relay selection, we introduce virtual relays representing subchannels, transforming the many-to-one bipartite matching problem into a one-to-one matching problem, thereby enhancing computational efficiency. To handle content popularity uncertainties, we incorporate the notion of “gamma-robustness” into the b-matching framework, enabling optimization across multiple popularity scenarios. Moreover, to address channel estimation errors, we adopt Soyster’s worst-case uncertainty approach, ensuring feasible relay selection despite channel fluctuations. By integrating these robust optimization techniques, our framework mitigates performance degradation caused by unpredictable content popularity and channel conditions, thereby ensuring reliable and resilient network performance. The proposed method maintains polynomial time complexity, making it suitable for practical deployment. Through simulations, we evaluate its effectiveness by comparing it with learning-based solutions under uncertainty, demonstrating its superior performance.</p>