Language models for low resource text classification Large pretrained models can be used for text classification in zero shot conditions, but how to improve them to production level accuracy? A review of classical finetuning and prompt tuning.
Apr 23, 2022
Efficient offline search evaluation A discussion of search evaluation, with a system for test collection creation and management. Active ranking with pairwise comparisons is used to minimize assessor work, and partial relevance metrics for metrics calculation.