Existing online social network (OSN) services use caching systems with the least recently used (LRU) algorithm as an eviction policy for improving service performance. However, they do not consider the characteristics of users’ usage pattern in OSN services. In addition, they do not consider the fact that the users and cloud servers are geographically distributed over a large area. It makes relatively unnecessary data occupy limited memory space. Consequently, they cannot prevent the degradation of cache efficiency. We introduce a social-aware caching algorithm to improve the performance of OSN services in a multi-cloud environment. Our approach is designed to consider the locations of the user and cloud server and to allocate memory space differently to each user by considering the user’s frequency of service usage. To validate our approach, we implemented a OSN service that manages user data in the same way as Twitter that is a representative OSN service. Furthermore, we experimented with actual users’ locations and times of use as collected from Twitter. Our findings indicate that this approach can improve the cache hit ratio by an average of more than 24% and reduce the execution delay by an average of more than 1095 ms.