Reproduce “Compositional Generalization for Multi-label Text Classification: A Data-Augmentation Approach” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuyang Chai, Zhuang Li, Jiahui Liu, Lei Chen, Fei Li, Donghong Ji, Chong Teng

The authors test whether classifiers handle familiar labels in new combinations. LD-VAE generates and filters texts with label-specific representations and conditional priors for classifier training. On AAPD, exact-match accuracy rises from 21.96% without augmentation to 31.11% with LD-VAE augmentation. In three constructed splits, 20 combinations are held out, but support examples remain, so this tests composition shift, not zero-shot generalization.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.