Dense Retrieval as Indirect Supervision for Large-space Decision Making
This addresses the problem of handling large label spaces in NLU for researchers and practitioners, offering a novel approach with significant performance gains, though it builds on existing dense retrieval methods.
The paper tackles the challenge of learning large-space discriminative NLU tasks by reformulating them as a learning-to-retrieve task, resulting in Dense Decision Retrieval (DDR) which outperforms baselines by up to 27.54% in P@1 on extreme multi-label classification and improves F1 and accuracy on other tasks.
Many discriminative natural language understanding (NLU) tasks have large label spaces. Learning such a process of large-space decision making is particularly challenging due to the lack of training instances per label and the difficulty of selection among many fine-grained labels. Inspired by dense retrieval methods for passage finding in open-domain QA, we propose a reformulation of large-space discriminative NLU tasks as a learning-to-retrieve task, leading to a novel solution named Dense Decision Retrieval (DDR ). Instead of predicting fine-grained decisions as logits, DDR adopts a dual-encoder architecture that learns to predict by retrieving from a decision thesaurus. This approach not only leverages rich indirect supervision signals from easy-to-consume learning resources for dense retrieval, it also leads to enhanced prediction generalizability with a semantically meaningful representation of the large decision space. When evaluated on tasks with decision spaces ranging from hundreds to hundred-thousand scales, DDR outperforms strong baselines greatly by 27.54% in P@1 on two extreme multi-label classification tasks, 1.17% in F1 score ultra-fine entity typing, and 1.26% in accuracy on three few-shot intent classification tasks on average. Code and resources are available at https://github.com/luka-group/DDR