|
| 1 | +--- |
| 2 | +# Documentation: https://wowchemy.com/docs/managing-content/ |
| 3 | + |
| 4 | +title: "FitCF: A Framework for Automatic Feature Importance-guided Counterfactual Example Generation" |
| 5 | +authors: ["Qianli Wang", "Nils Feldhus", "Simon Ostermann", "Luis Felipe Villa-Arenas", "Sebastian Möller", "Vera Schmitt"] |
| 6 | +date: 2025-05-19T10:42:03+02:00 |
| 7 | +doi: "" |
| 8 | + |
| 9 | +# Schedule page publish date (NOT publication's date). |
| 10 | +publishDate: 2025-05-19T10:42:03+02:00 |
| 11 | + |
| 12 | +# Publication type. |
| 13 | +# Legend: 0 = Uncategorized; 1 = Conference paper; 2 = Journal article;# 3 = Preprint / Working Paper; 4 = Report; 5 = Book; 6 = Book section;# 7 = Thesis; 8 = Patent |
| 14 | +publication_types: ["1"] |
| 15 | + |
| 16 | +# Publication name and optional abbreviated publication name. |
| 17 | +publication: "The 63rd Annual Meeting of the Association for Computational Linguistics" |
| 18 | +publication_short: "ACL 2025 Findings" |
| 19 | + |
| 20 | +abstract: "Counterfactual examples are widely used in natural language processing (NLP) as valuable data to improve models, and in explainable artificial intelligence (XAI) to understand model behavior. The automated generation of counterfactual examples remains a challenging task even for large language models (LLMs), despite their impressive performance on many tasks. In this paper, we first introduce ZeroCF, a faithful approach for leveraging important words derived from feature attribution methods to generate counterfactual examples in a zero-shot setting. Second, we present a new framework, FitCF, which further verifies aforementioned counterfactuals by label flip verification and then inserts them as demonstrations for few-shot prompting, outperforming two state-of-the-art baselines. Through ablation studies, we identify the importance of each of FitCF's core components in improving the quality of counterfactuals, as assessed through flip rate, perplexity, and similarity measures. Furthermore, we show the effectiveness of LIME and Integrated Gradients as backbone attribution methods for FitCF and find that the number of demonstrations has the largest effect on performance. Finally, we reveal a strong correlation between the faithfulness of feature attribution scores and the quality of generated counterfactuals." |
| 21 | + |
| 22 | +# Summary. An optional shortened abstract. |
| 23 | +summary: "" |
| 24 | + |
| 25 | +tags: [] |
| 26 | +categories: [] |
| 27 | +featured: false |
| 28 | + |
| 29 | +# Custom links (optional). |
| 30 | +# Uncomment and edit lines below to show custom links. |
| 31 | +# links: |
| 32 | +# - name: Follow |
| 33 | +# url: https://twitter.com |
| 34 | +# icon_pack: fab |
| 35 | +# icon: twitter |
| 36 | + |
| 37 | +url_pdf: "https://arxiv.org/abs/2501.00777" |
| 38 | +url_code: "" |
| 39 | +url_dataset: "" |
| 40 | +url_poster: |
| 41 | +url_project: |
| 42 | +url_slides: |
| 43 | +url_source: |
| 44 | +url_video: |
| 45 | + |
| 46 | +# Featured image |
| 47 | +# To use, add an image named `featured.jpg/png` to your page's folder. |
| 48 | +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. |
| 49 | +image: |
| 50 | + caption: "" |
| 51 | + focal_point: "" |
| 52 | + preview_only: false |
| 53 | + |
| 54 | +# Associated Projects (optional). |
| 55 | +# Associate this publication with one or more of your projects. |
| 56 | +# Simply enter your project's folder or file name without extension. |
| 57 | +# E.g. `internal-project` references `content/project/internal-project/index.md`. |
| 58 | +# Otherwise, set `projects: []`. |
| 59 | +projects: [VERANDA] |
| 60 | + |
| 61 | +# Slides (optional). |
| 62 | +# Associate this publication with Markdown slides. |
| 63 | +# Simply enter your slide deck's filename without extension. |
| 64 | +# E.g. `slides: "example"` references `content/slides/example/index.md`. |
| 65 | +# Otherwise, set `slides: ""`. |
| 66 | +slides: "" |
| 67 | +--- |
0 commit comments