User reputation is a crucial indicator in social networks, where it is exploited to promote authoritative content and to marginalize spammers. To be accurate, reputation must be updated periodically, taking into account the whole historical data of user activity. In big social networks like Twitter and Facebook, these updates would require to process a huge amount of historical data, and therefore pose serious performance issues.
We address these issues in the context of Twitter, by studying a technique which can update user reputation in constant time. This is obtained by using an arbitrary ranking algorithm to compute user reputation in the most recent time window, and by combining it with a summary of historical data. Experimental evaluation on large datasets show that our technique improves the performance of existing ranking algorithms, at the cost of a negligible degradation of their precision.
Faderank sources: https://bitbucket.org/tcsunicait/twitter-reputation
Validation data: faderank-validation.zip
Note: datasets are not available online, because not permitted by I.6.b of the Twitter Developer Agreement