Ридж

We introduce three linguistically moti- vated structured regularizers based on parse trees, topics, and hierarchical word clusters for text categorization. These regularizers impose linguistic bias in fea- ture weights, enabling …...