The present dataset is designed for a distributed broker-based system in which user profiles are hosted by pub/sub separate brokers and each broker subscribes to the activities performed by friends of the users that it hosts (on their behalf). The dataset parameters are as follows:
- number of brokers (broker-no): 120
- number of users (user-no): 350,000
- number of subscriptions (subs-no): 125,612
- number of publication (pubs-no): 50,024
- number of deliveries: 1.22 million
The compilation process is as follows:
- A subset of user-no of unique users has been randomly picked and the social graph (friendships) and interaction graph (wall posts) present in the Facebook traces has been contracted to include these users only.
- The selected users were assigned uniformly between brokers.
- Subscriptions local to each broker were then constructed according to the identifiers of friends of each user assigned to that broker (based on social graph traces).
- Publications are then constructed based on the interaction graphs such that each publication matches friends of the user who wrote the wall post as well as the friends of the user who received the post.
The number of hosts/users can be configured and the pub/sub dataset be recompiled accordingly from the original user traces using our scripts. However, due to privacy issues we are unable to distribute original user traces and scripts for recompiling the dataset. If you wish access to a new dataset with different parameters please contact us with the specifics of your request. We try our best to accommodate your requests.
References: Christo Wilson, Bryce Boe, Alessandra Sala, Krishna P.N. Puttaswamy, and Ben Y. Zhao. 2009. User interactions in social networks and their implications. In Proceedings of the 4th ACM European conference on Computer systems (EuroSys '09). ACM, New York, NY, USA, 205-218.