Add paper link, GitHub link, and task categories
#2
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,70 +1,75 @@
|
|
| 1 |
-
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
|
| 37 |
-
|
|
| 38 |
-
| `
|
| 39 |
-
| `
|
| 40 |
-
| `
|
| 41 |
-
| `
|
| 42 |
-
| `
|
| 43 |
-
| `
|
| 44 |
-
| `
|
| 45 |
-
| `
|
| 46 |
-
| `
|
| 47 |
-
| `
|
| 48 |
-
| `
|
| 49 |
-
| `
|
| 50 |
-
| `
|
| 51 |
-
| `
|
| 52 |
-
| `
|
| 53 |
-
| `
|
| 54 |
-
| `
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
-
|
| 67 |
-
-
|
| 68 |
-
|
| 69 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
The end result is a filtered rejected subset, a filtered accepted subset, and a final accepted sample that is more balanced for downstream analysis.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
pretty_name: ACL ARR Reviews - Review Arcade Project
|
| 4 |
+
task_categories:
|
| 5 |
+
- text-classification
|
| 6 |
+
- text-generation
|
| 7 |
+
configs:
|
| 8 |
+
- config_name: acl_arr_reviews
|
| 9 |
+
data_files:
|
| 10 |
+
- split: accepted
|
| 11 |
+
path: data/accepted/*
|
| 12 |
+
- split: rejected
|
| 13 |
+
path: data/rejected/*
|
| 14 |
+
- split: accepted_final
|
| 15 |
+
path: data/accepted_final/*
|
| 16 |
+
- split: rejected_final
|
| 17 |
+
path: data/rejected_final/*
|
| 18 |
+
- split: rejected_short
|
| 19 |
+
path: data/rejected_short/*
|
| 20 |
+
- split: accepted_short
|
| 21 |
+
path: data/accepted_short/*
|
| 22 |
+
---
|
| 23 |
+
|
| 24 |
+
# ACL ARR Reviews - Review Arcade Project
|
| 25 |
+
|
| 26 |
+
This dataset contains paper reviews from the ACL ARR (Association for Computational Linguistics - Annual Review of Research) program. The reviews are organized into splits corresponding to papers that were accepted or rejected for publication, as well as specific subsets used for the research analysis.
|
| 27 |
+
|
| 28 |
+
This dataset was introduced in the paper [Review Arcade: On the Human Alignment and Gameability of LLM Reviews](https://huggingface.co/papers/2605.28897).
|
| 29 |
+
|
| 30 |
+
**Code**: [https://github.com/uhh-hcds/reviewarcade](https://github.com/uhh-hcds/reviewarcade)
|
| 31 |
+
|
| 32 |
+
## Record Format
|
| 33 |
+
|
| 34 |
+
Each line in the JSONL files represents one paper review record. The fields below describe the expected schema.
|
| 35 |
+
|
| 36 |
+
| Field | Type | Description |
|
| 37 |
+
| --- | --- | --- |
|
| 38 |
+
| `title` | string | Paper title. |
|
| 39 |
+
| `authors` | string | Author list as a single string. |
|
| 40 |
+
| `accepted_at` | string | Venue or event where the paper was accepted (for example `acl25`). |
|
| 41 |
+
| `abstract` | string | Paper abstract text. |
|
| 42 |
+
| `cycle` | string | Review cycle identifier/source path (for example `aclweb.org/ACL/ARR/2024/December`). |
|
| 43 |
+
| `mean_overall_assessment` | number | Mean of reviewer overall-assessment scores. |
|
| 44 |
+
| `overall_assessment` | array[number] | Raw reviewer overall-assessment scores. |
|
| 45 |
+
| `mean_confidence` | number | Mean of reviewer confidence scores. |
|
| 46 |
+
| `confidence` | array[number] | Raw reviewer confidence scores. |
|
| 47 |
+
| `paper_summary` | array[string] | Reviewer-written summaries of the paper. |
|
| 48 |
+
| `summary_of_strengths` | array[string] | Reviewer-listed strengths. |
|
| 49 |
+
| `summary_of_weaknesses` | array[string] | Reviewer-listed weaknesses. |
|
| 50 |
+
| `comments_suggestions_and_typos` | array[string] | Reviewer comments, suggestions, and typo/edit notes. |
|
| 51 |
+
| `soundness` | array[number] | Reviewer soundness scores. |
|
| 52 |
+
| `excitement` | array[number] | Reviewer excitement scores (can be empty). |
|
| 53 |
+
| `datasets` | array[number] | Reviewer dataset-related scores. |
|
| 54 |
+
| `software` | array[number] | Reviewer software/reproducibility scores. |
|
| 55 |
+
| `metareview_score` | number | Meta-review score. |
|
| 56 |
+
| `metareview` | string | Meta-review summary text. |
|
| 57 |
+
| `summary_of_reasons_to_publish` | string | Meta-review rationale for publication. |
|
| 58 |
+
| `text_v1` | string | Full paper/review text snapshot (version 1). |
|
| 59 |
+
| `text_v2` | string | Full paper/review text snapshot (version 2, camera-ready version). |
|
| 60 |
+
|
| 61 |
+
## Reviewed Subset Workflow
|
| 62 |
+
|
| 63 |
+
The reviewed subsets are built to focus the analysis on papers with actual reviewer feedback:
|
| 64 |
+
|
| 65 |
+
- Papers with no review entries are left out of the reviewed subsets.
|
| 66 |
+
- This creates a cleaner comparison between accepted and rejected papers, since both groups now contain only papers that were evaluated by reviewers.
|
| 67 |
+
- The accepted side is kept free of duplicate paper entries so each paper is represented once in the final sample.
|
| 68 |
+
|
| 69 |
+
The final accepted sample is designed to favor papers with richer review coverage while still filling the full target size of 1392:
|
| 70 |
+
|
| 71 |
+
- Papers with multiple reviews are prioritized first.
|
| 72 |
+
- Papers with a single review are used next.
|
| 73 |
+
- Any remaining space is filled from the rest of the reviewed accepted papers.
|
| 74 |
+
|
| 75 |
The end result is a filtered rejected subset, a filtered accepted subset, and a final accepted sample that is more balanced for downstream analysis.
|