Upload SurveyReview v1.0-paper data

.gitattributes

@@ -58,3 +58,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+v1.0-paper/articles/articles_part1.json filter=lfs diff=lfs merge=lfs -text
+v1.0-paper/articles/articles_part2.json filter=lfs diff=lfs merge=lfs -text
+v1.0-paper/raw/train_samples.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,52 @@
+---
+pretty_name: SurveyReview
+language:
+- en
+tags:
+- survey-evaluation
+- peer-review
+- benchmark
+- text
+task_categories:
+- text-classification
+- text-generation
+---
+
+# SurveyReview
+
+SurveyReview is a reviewer-aligned benchmark for survey evaluation.
+
+This repository contains the `v1.0-paper` release, which is aligned with the data statistics reported in the paper.
+
+## Data Layout
+
+- `v1.0-paper/raw/`: review-level train/test samples matching the paper statistics.
+- `v1.0-paper/articles/`: article data used in the paper experiments.
+- `v1.0-paper/prompt/`: prompts used by the evaluation pipeline.
+- `v1.0-paper/train/`: survey-grouped train data for direct use.
+- `v1.0-paper/test/`: survey-grouped test data for direct use.
+
+## Splits
+
+| Split | Count |
+| --- | ---: |
+| Raw train samples | 1,216 |
+| Raw test samples | 414 |
+| Grouped train surveys | 480 |
+| Grouped test surveys | 163 |
+
+## Usage
+
+```python
+from huggingface_hub import snapshot_download
+
+local_dir = snapshot_download(
+    repo_id="brighterrluo/SurveyReview",
+    repo_type="dataset",
+)
+print(local_dir)
+```
+
+## Version
+
+`v1.0-paper` is the paper-aligned version for reproducing the benchmark setting.

v1.0-paper/articles/articles_part1.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6688ed4854ff1b06fcbc87315d91e99ff06cca8dc790f12ed18e630a3535a952
+size 95898492

v1.0-paper/articles/articles_part2.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f1aa9a2971a4e67cc57eae4f425eeb64df32f5970c5c488eaf9c2558d2d23cb
+size 95899449

v1.0-paper/prompt/Definition.json

@@ -0,0 +1,7 @@
+{
+  "Readability": "Readability refers to the manuscript's overall presentation, focusing on clarity and ease of understanding.",
+  "Criticalness": "Criticalness refers to the manuscript's originality and depth, focusing on whether it offers new insights, critical perspectives, or meaningful contributions beyond existing knowledge.",
+  "Comprehensiveness": "Comprehensiveness refers to how complete and relevant the manuscript's references are, considering whether key works are cited, outdated or irrelevant sources are avoided.",
+  "Structure": "Structure refers to the logical organization and coherence of sections and subsections, ensuring that they are logically connected."
+}
+

v1.0-paper/prompt/eval-prompt.json

@@ -0,0 +1,7 @@
+{
+  "Readability": "Please act as a research paper reviewer and evaluate the following manuscript excerpt from the perspective of readability.\nReadability refers to the manuscript's overall presentation, focusing on clarity and ease of understanding.\n### Provide the following:\nReasoning (briefly explain your evaluation)\nReadability Score (choose one of: -2, -1, 1, 2) \n### Scoring Guide:\n-2 = Very hard to understand; poor structure and clarity \n-1 = Somewhat unclear; ideas or structure need improvement \n1 = Basically clear; statements are understandable \n2 = Very clear, concise, and well-presented throughout \n### Output Format:\nReason$$$X$$$\n#### Example:\nThe writing is fluent and well-organized, with clear logical flow throughout.$$$2$$$",
+  "Criticalness": "Please act as a research paper reviewer and evaluate the following manuscript excerpt from the perspective of criticalness.\nCriticalness refers to the manuscript's originality and depth, focusing on whether it offers new insights, critical perspectives, or meaningful contributions beyond existing knowledge.\n### Provide the following:\nReasoning (briefly explain your evaluation)\nCriticalness Score (choose one of: -2, -1, 1, 2) \n### Scoring Guide:\n-2 = Lacks originality; offers no new insights or contributions \n-1 = Limited novelty; mostly reiterates existing knowledge \n1 = Shows reasonable originality; provides some new perspectives \n2 = Highly original and insightful; offers significant new contributions \n### Output Format:\nReason$$$X$$$\n#### Example:\nThe manuscript presents novel insights and critical perspectives that significantly advance the field.$$$2$$$",
+  "Comprehensiveness": "Please act as a research paper reviewer and evaluate the following manuscript excerpt from the perspective of comprehensiveness.\nComprehensiveness refers to how complete and relevant the manuscript's references are, considering whether key works are cited, outdated or irrelevant sources are avoided.\n### Provide the following:\nReasoning (briefly explain your evaluation)\nComprehensiveness Score (choose one of: -2, -1, 1, 2) \n### Scoring Guide:\n-2 = Major gaps in references; missing key works or many irrelevant sources \n-1 = Some important references missing; includes some outdated or irrelevant sources \n1 = Adequately cited; covers most key works with few gaps \n2 = Comprehensive and highly relevant; all key works cited appropriately \n### Output Format:\nReason$$$X$$$\n#### Example:\nThe references are comprehensive and up-to-date, covering all major works in the field.$$$2$$$",
+  "Structure": "Please act as a research paper reviewer and evaluate the following manuscript excerpt from the perspective of structure.\nStructure refers to the logical organization and coherence of sections and subsections, ensuring that they are logically connected.\n### Provide the following:\nReasoning (briefly explain your evaluation)\nStructure Score (choose one of: -2, -1, 1, 2) \n### Scoring Guide:\n-2 = Poorly organized; sections lack logical flow and coherence \n-1 = Some organizational issues; connections between sections unclear \n1 = Reasonably well-organized; sections mostly follow logical order \n2 = Excellently structured; sections are logically connected and flow seamlessly \n### Output Format:\nReason$$$X$$$\n#### Example:\nThe manuscript is well-structured with clear logical connections between all sections.$$$2$$$"
+}
+

v1.0-paper/prompt/reason_quality_judge.json

@@ -0,0 +1,3 @@
+{
+  "reason_quality_judge": "You are an expert evaluator tasked with assessing the quality of AI-generated review reasoning by comparing it against reference reasoning.\n\n### Your Task:\nEvaluate how well the predicted reasoning aligns with the reference reasoning for a given research paper review.\n\n### Input Information:\nYou will receive the following information:\n1. **Dimension**: The evaluation aspect (e.g., Criticalness, Readability, Comprehensiveness, Structure)\n2. **Dimension Definition**: The detailed definition of what this dimension measures\n3. **Paper Title**: The title of the paper being reviewed\n4. **Reference Reasoning**: The ground truth reasoning provided by human reviewers (true_reasons)\n5. **Predicted Output**: The reasoning generated by the AI model (pred_output)\n\n### Scoring Scale:\nYou must output ONLY ONE of the following scores:\n\n**1.0 - Perfect Alignment**\n- Core arguments are identical\n- All key details are covered\n- Logical structure matches perfectly\n- Only minor differences in wording\n\n**0.8 - High Alignment**\n- Core arguments are consistent\n- Most key details are covered\n- Similar logical flow\n- May have stylistic differences but meaning is the same\n\n**0.6 - Basic Alignment**\n- Main points are consistent\n- Some key details are covered\n- Logical direction is similar\n- Some omissions or minor deviations present\n\n**0.4 - Partial Alignment**\n- Some common points exist\n- Significant omissions or inconsistencies\n- Logical approach may differ\n- Only partially captures the reference reasoning\n\n**0.2 - Minimal Alignment**\n- Few common points\n- Major differences in arguments\n- Different focus or perspective\n- Mostly fails to match reference reasoning\n\n**0.0 - No Alignment**\n- Completely different arguments\n- No substantial overlap\n- Contradictory or irrelevant content\n- Fails to address the same aspects\n\n### Output Format:\nOutput ONLY a single number from the following options: 0.0, 0.2, 0.4, 0.6, 0.8, 1.0\n\n### Examples:\n\n#### Example 1 (Score: 1.0):\n**Reference**: \"The manuscript provides comprehensive coverage of recent advances with particular strength in discussing transformer architectures.\"\n**Predicted**: \"The survey excels in covering state-of-the-art approaches, especially in its thorough treatment of transformer models.\"\n**Score**: 1.0\n\n#### Example 2 (Score: 0.6):\n**Reference**: \"The discussion section identifies important gaps and provides valuable insights into future research directions.\"\n**Predicted**: \"The paper discusses some gaps in the literature and mentions potential future work.\"\n**Score**: 0.6\n\n#### Example 3 (Score: 0.2):\n**Reference**: \"The manuscript demonstrates strong originality and offers novel insights that advance the field.\"\n**Predicted**: \"The paper has methodological flaws and lacks proper analysis.\"\n**Score**: 0.2\n\n### Important Notes:\n- Focus on semantic meaning, not exact word matching\n- Consider the context provided by the dimension definition\n- Evaluate fairly - partial credit for partial alignment\n- Output ONLY the numerical score (0.0, 0.2, 0.4, 0.6, 0.8, or 1.0)"
+}

v1.0-paper/raw/train_samples.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c687a734425a8b0a5074aaab13ce1a0fd65eb2cff966d669c377a3562580fd03
+size 13167924