As wider use of digital camera in these decades, photograph data from individuals increases dramatically. Many photos with different people are available on the Internet. It stimulates a strong demand on automatic face annotation. Moreover, it becomes more possible to discover potential social information from increasingly large photo collections. Every photo in a photo collection is not isolated. Instead, they are highly related as a whole to represent an event, such as a wedding. In a particular event, people would appear as a group following some rules, like families show up in a wedding and colleagues from the same research group in a conference. We also found that clues of closeness between people imply in photos as well. This paper explores social community from personal photo collection with modularity and proposes a method combining ensemble RBFNN with pairwise social relationship as context for recognizing people. Experiments on a conference photo album shows that a certain embedded social network with community structure is revealed. Our simple approach of face recognition with social context enhances the annotation performance when compared with the baseline method.