WTF, GPU! computing twitter's who-to-follow on the GPU

In this paper, we investigate the potential of GPUs for performing link structure analysis of social graphs. Specifically, we implement Twitter's WTF ("Who to Follow") recommendation system on a single GPU. Our implementation shows promising results on moderate-sized social graphs. It can return the top-K relevant users for a single user in 172 ms when running on a subset of the 2009 Twitter follow graph with 16 million users and 85 million social relations. For our largest dataset, which contains 75% of the users (30 million) and 50% of the social relations (680 million) of the complete follow graph, this calculation takes 1.0 s. We also propose possible solutions to apply our system to follow graphs of larger sizes that do not fit into the on-board memory of a single GPU.