Joint Caching and Recommendation Optimization from Network and User Perspectives in Wireless D2D Networks

This work examines the impact of recommendation on both the shaping of user preference and the caching decisions at wireless edge devices. While most studies focus on the optimization from a network perspective, we investigate the joint caching and recommendation optimization for wireless device-to-device (D2D) networks from both network and user perspectives and identify their key differences. To achieve this goal, optimization problems for network offloading and user offloading probabilities as well as their tradeoff are first formulated. Two types of preference-shaping models are considered. The first type assumes that the user preference can be arbitrarily shaped whereas the second type only selectively enhances the users' original preference. The proposed optimization problems are solved using alternating optimization where the users' caching variables and the recommendation variables are optimized in turn until convergence. The corresponding convergence and complexity analyses are also provided. Extensive simulations are conducted to validate the efficacy of the solution approaches. Results show that designs obtained from different perspectives can lead to very different performance behaviors, and the difference is especially large when there is little similarity among users' preferences.