Content caching is widely recognized a promising functionality to improve service performance in mobile edge computing (MEC). In the big data era, there are massive heterogeneous contents collected by the mobile devices, belonging to different users with specific context (e.g., hobby, environment, age, etc). However, local content caching without content popularity and context information in advance is not accurate enough. Especially, multiple large-scale contents cached in the local database bring high pressure to the process of content selection. Hence, to handle these important issues, we propose a context-aware distributed online learning algorithm for efficient content caching according to a novel tree-based and contextual multi-arm bandit theory for collaborative MEC in this paper. To guarantee the trustworthy collaboration, we introduce a trust evaluation factor to find reliable neighboring ENs. Moreover, our system extracts contextual information from users into the context space and builds up a content cover tree to maximize caching hit rates to satisfy users’ demands. Our simulation results based on a real-world dataset indicate that our proposal can achieve a balance between caching hit rates and time cost, and have a sublinear bound of cumulative regret. This verifies its superior caching-hits performance gain compared to the other related algorithms.

中文翻译：

---

具有自动缩放功能的可信赖和上下文感知分布式在线学习，用于协作移动边缘计算中的内容缓存

内容缓存被广泛认为是提高移动边缘计算 (MEC) 服务性能的一种很有前途的功能。在大数据时代，移动设备收集到的海量异构内容，属于具有特定语境（如爱好、环境、年龄等）的不同用户。然而，没有预先内容流行度和上下文信息的本地内容缓存是不够准确的。尤其是本地数据库中缓存了多个大型内容，给内容选择的过程带来了很大的压力。因此，为了处理这些重要问题，我们在本文中提出了一种上下文感知分布式在线学习算法，用于根据一种新颖的基于树和上下文的多臂老虎机理论进行协作 MEC，用于有效的内容缓存。为保证可信赖的合作，我们引入了一个信任评估因子来寻找可靠的相邻 EN。此外，我们的系统将用户的上下文信息提取到上下文空间中，并构建内容覆盖树以最大化缓存命中率以满足用户的需求。我们基于真实世界数据集的模拟结果表明，我们的提议可以在缓存命中率和时间成本之间取得平衡，并且具有累积后悔的亚线性界限。与其他相关算法相比，这验证了其卓越的缓存命中性能增益。我们基于真实世界数据集的模拟结果表明，我们的提议可以在缓存命中率和时间成本之间取得平衡，并且具有累积后悔的亚线性界限。与其他相关算法相比，这验证了其卓越的缓存命中性能增益。我们基于真实世界数据集的模拟结果表明，我们的提议可以在缓存命中率和时间成本之间取得平衡，并且具有累积后悔的亚线性界限。与其他相关算法相比，这验证了其卓越的缓存命中性能增益。