Research map / Label-efficient and robust learning / Recognition and long-tail
Robust Text Classification in the Presence of Confounding Bias
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add back-door adjustment (confounder features at training, sum out at test) with strength v1=10 | L2 logistic regression · Twitter, large train/test confounder mismatch | accuracy: 60% -> 81% | Helped |
| add back-door adjustment BAZ10 when train and test confounding match | L2 logistic regression · Twitter / IMDb, small correlation difference | accuracy: approximately 2 points loss | Hurt |
| increase adjustment strength v1 from below 0.1 to 10 (ablation) | Back-door adjustment · Twitter, large bias difference | accuracy: 15 points between plateaus | Helped |
| add back-door adjustment BAZ10 vs LR across correlation differences | L2 logistic regression · Twitter and IMDb | accuracy (paired t-test): 19 wins, 8 losses, 5 ties | Mixed |
| use back-door adjustment (sum out) instead of modeling joint P(Y,Z|X) and summing out | Sum out classifier · Twitter | accuracy: SO is 4 to 8 points lower overall | Helped |
| compare subsampling (LRS) with back-door adjustment when label/confounder joint distribution is skewed | Subsampling · IMDb | accuracy | Hurt |
| add back-door adjustment BAZ10 after confounder flips | L2 logistic regression · Canadian Parliament | accuracy | Helped |
| subsample to discard pre-shift data (LRS) vs BAZ10 with only 5 post-shift instances | Back-door adjustment · Canadian Parliament | accuracy | Hurt |
Closest papers by meaning
- Robustness to Spurious Correlations in Text Classification via Automatically Generated Counterfactuals (AAAI 2021 · similarity 0.88)
- Causal Estimation for Text Data with (Apparent) Overlap Violations (ICLR 2023 · similarity 0.85)
- Learning The Difference That Makes A Difference With Counterfactually-Augmented Data (ICLR 2020 · similarity 0.84)
- Leveraging Uncertainty Estimates To Improve Classifier Performance (ICLR 2024 · similarity 0.81)
- Data-faithful Feature Attribution: Mitigating Unobservable Confounders via Instrumental Variables (NeurIPS 2024 · similarity 0.81)
- Adjustment for Confounding using Pre-Trained Representations (ICML 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.