Meta Algorithm

We address the rating-inference problem, wherein rather than simply decide whether a review is “thumbs up” or “thumbs down”, as in previous sentiment analy- sis work, one must determine an …...

We propose a new framework for N-best reranking on sparse feature sets. The idea is to reformulate the reranking problem as a Multitask Learning problem, where each N-best list corresponds …...

AbstractStructured prediction is often approached by training a locally normalized model with maximum likelihood and decoding approximately with beam search. This approach leads to mismatches as, during training, the model …...