CLMay 28, 2023

Targeted Data Generation: Finding and Fixing Model Weaknesses

arXiv:2305.17804v1233 citations
Originality Highly original
AI Analysis

This addresses fairness and trust issues in NLP by fixing model weaknesses for underrepresented subgroups, representing a novel method for a known bottleneck.

The paper tackles the problem of systematic failures in NLP models on specific subgroups despite high aggregate accuracy, proposing Targeted Data Generation (TDG) to automatically identify and generate data for these subgroups, which significantly improves accuracy on challenging subgroups and overall test accuracy in experiments.

Even when aggregate accuracy is high, state-of-the-art NLP models often fail systematically on specific subgroups of data, resulting in unfair outcomes and eroding user trust. Additional data collection may not help in addressing these weaknesses, as such challenging subgroups may be unknown to users, and underrepresented in the existing and new data. We propose Targeted Data Generation (TDG), a framework that automatically identifies challenging subgroups, and generates new data for those subgroups using large language models (LLMs) with a human in the loop. TDG estimates the expected benefit and potential harm of data augmentation for each subgroup, and selects the ones most likely to improve within group performance without hurting overall performance. In our experiments, TDG significantly improves the accuracy on challenging subgroups for state-of-the-art sentiment analysis and natural language inference models, while also improving overall test accuracy.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes