Learning to Predict Web Collaborations

Much of the knowledge available on the web today comes as a result of fruitful collaborations among large groups of people. One of the most striking examples of successful web collaboration is the online encyclopedia Wikipedia. The web is used as a collaboration platform by highly specialized blogging communities and by the scientific community. An important reason for the richness of content generated through web collaborations is that the participants in such collaborations are not constrained by geographic location. Thus, like-minded individuals from across the world can join their efforts. This also means, however, that web collaborators often do not know each other, and, thus, finding collaborators on the web is more difficult than it is with more traditional forms of collaboration that are initiated based on acquaintance. This difficulty is further exacerbated by the fact that web collaborations tend to be more dynamic as participants join and abandon communities. We consider the task of recommending project-specific potential collaborators to web users and propose an approach that is based on statistical relational learning. Our proposed model thus has the advantages that it can include complex features composed of multiple properties and relationships of the entities, it can handle the high levels of noise and uncertainty inherent in user actions, and it allows for joint decision-making, which leads to more accurate predictions. To ensure scalability, our model is trained in an online fashion. We demonstrate the effectiveness of our approach on a data set collected from Wikipedia.