All Facebook profiles show mostly similar behaviors in the friend list. Our friends mostly do similar things that we liked or they think mostly in the same way. This is the basic principle of “all suggested things”
When you start to create a facebook page ads, you will have two options as; people who like your page, friends of liked people. If someone liked your page, his friends will care for your page by greater probability.
User Engagement.
Participation in a social system such as Facebook is built upon a spectrum of social decisions, beginning with the decision to join (recruitment) and continuing on to decisions about how to choose a level of engagement. We now show how structural diversity also plays an analogous role in this latter type of decision process, studying long-term user engagement in the Facebook service. Whereas recruitment is a function of the complex interplay between multiple acts of endorsement, engagement is a function of the social utility a user derives from the service. Our study of engagement focuses on users who registered for Facebook during 2010, analyzing the diversity of their social neighborhoods 1 week after registration as a basis for predicting whether they will become highly engaged users 3 months later.
- The logic is the same in any suggestion things. There are special algorithms for facebook but i will focus the common algorithms of facebook with other social networks.
Clustering Algorithms for Suggested and Similiar Mass-Generating
- Markov Clustering (MCL)(a): This algorithm is a disjoint clustering algorithm that uses the concept of Markov chains to simulate stochastic flows in graphs and builds a fast and scalable unsupervised clustering algorithm. MCL has a relatively high performance and is scalable.
- OSLOM(b): The Order Statistics Local Optimization Method (OSLOM) is an overlapping clustering algorithm that is among the first to account for edge weights and overlapping groups. It has a high performance and is scalable to large networks.
- Louvain(c): This hierarchical clustering algorithm uses modularity as its objective function and maximizes it using multiple heuristics to detect the groups. While this algorithm finds groups in a hierarchical manner, the lowest level of the hierarchies, which are the subgroups, are disjoint; i.e. one person cannot be a member of more than one group at the same level. The Louvain algorithm is highly accurate and has a very low computation time.
Algorithm system has same logical techniques, you can review a friendship map below:
