Mitigating Word Bias in Zero-shot Prompt-based Classifiers

TL;DR

This paper proposes an unsupervised method to mitigate word bias in zero-shot prompt classifiers by reweighting class probabilities to match uniform priors, improving performance across NLP tasks.

Contribution

It introduces a novel approach linking class priors to language model word priors, enabling threshold setting without labeled data, and demonstrates significant performance gains.

Findings

Matching class priors correlates with upper bound performance

Unsupervised reweighting improves prompt classifier accuracy

Method achieves consistent gains across multiple NLP tasks

Abstract

Prompt-based classifiers are an attractive approach for zero-shot classification. However, the precise choice of the prompt template and label words can largely influence performance, with semantically equivalent settings often showing notable performance difference. This discrepancy can be partly attributed to word biases, where the classifier may be biased towards classes. To address this problem, it is possible to optimise classification thresholds on a labelled data set, however, this mitigates some of the advantages of prompt-based classifiers. This paper instead approaches this problem by examining the expected marginal probabilities of the classes. Here, probabilities are reweighted to have a uniform prior over classes, in an unsupervised fashion. Further, we draw a theoretical connection between the class priors and the language models' word prior, and offer the ability to set a threshold in a zero-resource fashion. We show that matching class priors correlates strongly with the oracle upper bound performance and demonstrate large consistent performance gains for prompt settings over a range of NLP tasks.