Social media has become a global phenomenon. Currently, there are 2 billion active users on Facebook. However, much of the research on social media is about the consumption side of social media rather than the production or operational aspects of social media. Although research on the production side is still relatively small, it is growing, indicating that it is a fruitful area to study. This thesis attempts to contribute to this area of research to unravel the inner operations of social media with one key research question: How does social media platform organize information? The theory of digital object of Kallinikos et al. (2013) is used to investigate this question. Information display that users of a social media platform interact with is a digital object and it is constructed by two key components which are a database and algorithms. The database and the algorithms shape how information is being organized on information displays, and these influence user behaviors which are then captured as social data in the database. This thesis also critically examines the technology of recommender system by importing engineering literature on information filtering and retrieval. While newsfeed algorithm such as EdgeRank of Facebook has already been critically examined, information systems and media scholars have yet to investigate recommendation algorithms, despite the fact that they have been widely deployed all over the Internet. It is found that the key weakness of recommendation algorithms is their inability to recommend novel items. This is because the main tenet of any recommender system is to “recommend similar items to those that users already like”. Fortunately, this problem can be alleviated when recommender system is being deployed in the digital information environment of social media platforms. In turn, seven theoretical conjectures can be postulated. These are (1) navigation of information display as assembled by social media is highly interactive, (2) information organization of social media is highly unstable which would also render user behaviors unstable, (3) quality of data aggregation casts significant implications on user behaviors, (4) the amount of data captured by social media platforms limits the usefulness of their information displays, (5) output from the recommendation algorithm (recommendation list) casts real implications on user behaviors, (6) circle of friends on a social network can influence user behaviors, and (7) metadata attached to items being displayed casts influence on user behaviors. Data from Last.fm, a social media for music discovery, is used to evaluate these conjectures. The analysis supported most of the conjectures except the instability of information display and the importance of metadata attached to items being displayed. Some kinds of information organization are more stable than initially expected and some kinds of user generated contents are not so important for user behaviors.
Akarapat Charoenpanich (Sun,) studied this question.