Social media, born in this era of computers and online networking, provides a platform for users to build online social relations. Facebook, the most popular Online Social Network (OSN), could be used as a platform to share secret messages through JPEG images. However, due to the various lossy operations conducted over Facebook, the data embedded into the JPEG images can be easily destroyed, making the data extraction infeasible. More importantly, all these operations are carried out without users interference. In this thesis, we first perform an in-depth investigation of the various lossy operations that Facebook applies to uploaded images. Based upon such prior knowledge, we propose a DCT-domain data hiding scheme that can effectively embed a large amount of data and successfully extract them out from the downloaded images, defeating the uncontrolled lossy operations. Compared with the state-of-the-art techniques, the proposed method offers much higher embedding capacity, and can extract the data successfully with very high probability. Furthermore, the restored image upon data extraction is of high quality, and the file size expansion is negligible. Extensive experimental results are provided to validate our findings.