Proceedings Abstracts of the Twenty-Fifth International Joint Conference on Artificial Intelligence

Identifying Key Observers to Find Popular Information in Advance / 3761
Takuya Konishi, Tomoharu Iwata, Kohei Hayashi, Ken-ichi Kawarabayashi

Identifying soon-to-be-popular items in web services offers important benefits. We attempt to identify users who can find prospective popular items. Such visionary users are called observers. By adding observers to a favorite user list, they act to find popular items in advance. To identify efficient observers, we propose a feature selection based framework. This uses a classifier to predict item popularity, where the input features are a set of users who adopted an item before others. By training the classifier with sparse and non-negative constraints, observers are extracted as users whose parameters take a non-zero value. In experiments, we test our approach using real social bookmark datasets. The results demonstrate that our approach can find popular items in advance more effectively than baseline methods.