We study an approach to tweet classification based on distant supervision, whereby we automatically transfer labels from one social medium to another. In particular, we apply classes assigned to YouTube videos to tweets linking to these videos. This provides for free a virtually unlimited number of labelled instances that can be used as training data. The experiments we have run show that a tweet classifier trained via these automatically labelled data substantially outperforms an analogous classifier trained with a limited amount of manually labelled data.
|Title of host publication||Proceedings of the Ninth International Conference on Web and Social Media, ICWSM 2015, University of Oxford, Oxford, UK, May 26-29, 2015|
|Number of pages||4|
|Publication status||Published - Apr 2015|