SeedBERT: Recovering Annotator Rating Distributions from an Aggregated Label

TL;DR

SeedBERT is a novel method that recovers annotator rating distributions from single labels by inducing pre-trained models to attend to different input parts, improving performance on subjective tasks.

Contribution

It introduces SeedBERT, a technique that infers annotator disagreement distributions from one label, addressing the lack of annotator-specific data in subjective datasets.

Findings

SeedBERT's attention aligns with human annotator disagreement.

It outperforms standard models on subjective tasks.

Demonstrates significant performance gains in empirical evaluations.

Abstract

Many machine learning tasks -- particularly those in affective computing -- are inherently subjective. When asked to classify facial expressions or to rate an individual's attractiveness, humans may disagree with one another, and no single answer may be objectively correct. However, machine learning datasets commonly have just one "ground truth" label for each sample, so models trained on these labels may not perform well on tasks that are subjective in nature. Though allowing models to learn from the individual annotators' ratings may help, most datasets do not provide annotator-specific labels for each sample. To address this issue, we propose SeedBERT, a method for recovering annotator rating distributions from a single label by inducing pre-trained models to attend to different portions of the input. Our human evaluations indicate that SeedBERT's attention mechanism is consistent with human sources of annotator disagreement. Moreover, in our empirical evaluations using large language models, SeedBERT demonstrates substantial gains in performance on downstream subjective tasks compared both to standard deep learning models and to other current models that account explicitly for annotator disagreement.