Upload README.md with huggingface_hub

README.md

@@ -15,196 +15,240 @@ tags:
   - benchmark
 size_categories:
   - n<1K
+pretty_name: ResumeExtractBench
 ---
 
 # ResumeExtractBench
 
-A benchmark for evaluating schema-guided structured extraction from resume documents (PDFs). Contains **115 resumes** spanning expert-created templates, intentional parsing distractors, and handwritten documents.
+**ResumeExtractBench** is a benchmark for schema-guided structured extraction from resume documents. Given a resume PDF and a JSON Schema, systems must return structured data covering personal details, work history, education, skills, and more.
 
-## Overview
+- **Dataset Size**: 38 documents (handwritten + adversarial distractors)
+- **Schema Sections Scored**: 9 (basics, experience, education, projects, summary, certifications, awards, volunteering, skills)
+- **Domains**: 6 (engineering, healthcare, legal, data science, government, general)
+- **License**: CC-BY-4.0
 
-ResumeExtractBench evaluates how well extraction systems parse resumes into structured JSON data against a fixed schema. Each test case includes a source PDF, a JSON Schema defining the target output structure, and a human-verified ground truth extraction.
+**Quick Links**: [💻 Code & CLI](https://github.com/careerflow/resume-extract-bench) | [🏢 Careerflow](https://careerflow.ai)
 
-### Key Statistics
+---
+
+## Dataset Introduction
+
+### Document Composition
+
+| Challenge Category | Documents | Description |
+|---|---|---|
+| **Handwritten resumes** | 28 | Scanned handwritten resumes testing OCR and layout understanding |
+| **Adversarial distractors** | 10 | Synthetic resumes with intentional parsing challenges |
+| **Total** | **38** | |
+
+### Source Composition
 
-| Metric | Value |
-|--------|-------|
-| Total documents | 115 |
-| Handwritten resumes | 28 |
-| Adversarial distractor resumes | 10 |
-| LaTeX template resumes | 14 |
-| Expert-created (Canva) resumes | 63 |
-| Domains covered | 11 |
-| Schema sections scored | 9 |
+- **Handwritten documents**: 28
+- **Synthetic (LaTeX-generated)**: 10
+- **All documents classified as "hard" difficulty**
 
-### Difficulty Distribution
+### Challenge Categories
 
-| Difficulty | Count | Description |
-|------------|-------|-------------|
-| Hard | 38 | Handwritten + adversarial distractors |
-| Medium | 77 | Expert-created digital templates (Canva, LaTeX) |
+**C1: Handwritten Resumes** — Scanned handwritten documents with natural variation in handwriting, layout, and legibility. Tests OCR accuracy, spatial reasoning, and robustness to non-standard formatting. Typical failures: misread characters, merged/split words, missed sections, hallucinated content from ambiguous handwriting.
+
+**C2: Adversarial Distractors** — Synthetic resumes with intentional parsing challenges designed to stress-test extraction robustness:
+
+| Distractor | Challenge |
+|---|---|
+| **Date chaos** | Inconsistent date formats within the same resume (Jan 2020, 2020-06, 06/2021, Summer 2022) |
+| **Esoteric titles** | Non-standard job titles (Chief Vibes Officer, Growth Hacker), salary listed, entity confusion |
+| **European regional** | German-style CV with date of birth, marital status, nationality, military service, German grade scale |
+| **Functional format** | Skills-first layout with no chronological work history |
+| **Identity confusion** | Multi-part names, honorifics, maiden names, father's name, testimonial quotes with other people's titles |
+| **Multilingual** | English resume with French and Spanish intertwined |
+| **Prompt injection** | Hidden white text injection, PDF metadata injection, keyword stuffing in 6pt font |
+| **Prose narrative** | Third-person writing, no bullet points, 500-word personal statement, emoji section markers |
+| **Same company repeat** | Same employer listed 4 times with company renamed mid-tenure and overlapping dates |
+| **Skill-name collision** | Candidate name collides with programming language (Ruby Chen at Python Solutions Inc.) |
 
 ### Domain Coverage
 
-| Domain | Count |
-|--------|-------|
-| Engineering | 24 |
-| General | 30 |
-| Healthcare | 15 |
-| Data Science | 15 |
-| Legal | 12 |
-| Design | 8 |
-| Education | 4 |
-| Finance | 3 |
-| Software Engineering | 2 |
-| Consulting | 1 |
+| Domain | Documents |
+|---|---|
+| Engineering | 11 |
+| General | 10 |
+| Healthcare | 7 |
+| Legal | 7 |
+| Data Science | 2 |
 | Government | 1 |
 
-## Dataset Structure
+---
+
+## Usage
+
+### Loading with Datasets
+
+```python
+from huggingface_hub import snapshot_download
+import json
+
+root = snapshot_download(
+    repo_id="careerflow/ResumeExtractBench",
+    repo_type="dataset",
+)
+
+with open(f"{root}/test.jsonl") as f:
+    cases = [json.loads(line) for line in f if line.strip()]
 
+schema = json.load(open(f"{root}/schema.json"))
+
+for case in cases:
+    pdf_path = f"{root}/{case['files']['pdf']}"
+    ground_truth = case["ground_truth"]
+    # Run your extractor on pdf_path against schema
+    # Compare output to ground_truth
 ```
-ResumeExtractBench/
-├── README.md           # This file
-├── test.jsonl          # 115 test cases (one per line)
-├── schema.json         # Target JSON Schema for extraction
-└── pdfs/               # Source resume PDFs
-    ├── ai-engineer.pdf
-    ├── distractor-date-chaos-inconsistent-date-formats-throughout.pdf
-    ├── board-certified-ocularist-jane-doe.pdf
-    └── ...
+
+### Running Evaluation with CLI
+
+```bash
+pip install resume-extract-bench
+resume-bench download                                          # download dataset
+resume-bench run --pipeline gpt-5.6 --split test               # run extraction
+resume-bench grade --split test                                 # score against GT
+resume-bench leaderboard                                        # view results
 ```
 
-### JSONL Format
+### Bring Your Own Predictions
 
-Each line in `test.jsonl` is a JSON object:
+```bash
+resume-bench grade-file --predictions my_predictions.jsonl --split test
+```
+
+### Dataset Files
+
+- **test.jsonl**: 38 test cases (one JSON object per line)
+- **schema.json**: Target JSON Schema for extraction (resume_v1)
+- **pdfs/**: Source resume PDFs
+
+---
+
+## Dataset Format
+
+Each JSONL line represents one resume test case:
 
 ```json
 {
-  "resume_id": "distractor-date-chaos-inconsistent-date-formats-throughout",
-  "files": {"pdf": "pdfs/distractor-date-chaos-inconsistent-date-formats-throughout.pdf"},
-  "ground_truth": { ... },
-  "difficulty": "hard",
-  "layout_tags": ["distractor", "date-chaos"],
-  "source": "synthetic-distractor",
-  "domain": "software-engineering",
-  "schema_version": "resume_v1"
+    "resume_id": "distractor-date-chaos-inconsistent-date-formats-throughout",
+    "files": {"pdf": "pdfs/distractor-date-chaos-inconsistent-date-formats-throughout.pdf"},
+    "ground_truth": { ... },
+    "difficulty": "hard",
+    "layout_tags": ["distractor", "inconsistent-date-formats"],
+    "source": "synthetic-distractor",
+    "domain": "software-engineering",
+    "schema_version": "resume_v1"
 }
 ```
 
+### Field Definitions
+
 | Field | Type | Description |
-|-------|------|-------------|
-| `resume_id` | string | Unique identifier (filename stem) |
+|---|---|---|
+| `resume_id` | string | Unique identifier (PDF filename stem) |
 | `files.pdf` | string | Relative path to source PDF |
-| `ground_truth` | object | Human-verified extraction matching schema.json |
-| `difficulty` | string | `easy`, `medium`, or `hard` |
-| `layout_tags` | list | Visual/structural tags (e.g., `handwritten`, `distractor`, `canva`) |
-| `source` | string | Resume origin category |
-| `domain` | string | Professional domain |
+| `ground_truth` | object | Human-verified structured extraction conforming to `schema.json` |
+| `difficulty` | string | Difficulty level (`easy`, `medium`, `hard`) |
+| `layout_tags` | list[string] | Visual and structural challenge tags |
+| `source` | string | Resume origin: `expert-handwritten`, `synthetic-distractor` |
+| `domain` | string | Professional domain of the resume |
 | `schema_version` | string | Schema version (`resume_v1`) |
 
-### Schema (resume_v1)
+### Extraction Schema (resume_v1)
 
-The extraction schema defines 9 sections:
+The schema defines 9 sections that must be extracted from each resume:
 
-| Section | Type | Key Fields |
-|---------|------|------------|
-| `basics` | Singleton | fname, lname, email, phone, city, state, country, hasPersonalPhoto |
-| `experience` | Entity list | company, position, dates, city, description[] |
-| `education` | Entity list | institution, area, studyType, dates, description[] |
-| `projects` | Entity list | name, url, description[] |
-| `personalSummary` | Text | Free-text summary/objective |
-| `certifications` | Entity list | name, issuer |
-| `awards` | Entity list | title |
-| `volunteering` | Entity list | organization, position, dates, description[] |
-| `skills` | Flat list | category, skills[] |
+| Section | Type | Key Fields | Scored |
+|---|---|---|---|
+| `basics` | Singleton object | fname, lname, email, phone, city, state, country, hasPersonalPhoto | Per-field accuracy |
+| `experience` | Entity list | company, position, startMonth/Year, endMonth/Year, city, description[] | Entity P/R/F1 + description token F1 |
+| `education` | Entity list | institution, area, studyType, startMonth/Year, endMonth/Year, description[] | Entity P/R/F1 + description token F1 |
+| `projects` | Entity list | name, url, description[] | Entity P/R/F1 + description token F1 |
+| `personalSummary` | Free text | — | Token F1 |
+| `certifications` | Entity list | name, issuer | Entity P/R/F1 |
+| `awards` | Entity list | title | Entity P/R/F1 |
+| `volunteering` | Entity list | organization, position, startYear, endYear, description[] | Entity P/R/F1 + description token F1 |
+| `skills` | Flat list | category, skills[] | Set-level P/R/F1 |
 
 See `schema.json` for the full JSON Schema definition.
 
-## Resume Categories
+---
 
-### Adversarial Distractors (10 resumes)
+## Evaluation Metrics
 
-Synthetic resumes with intentional parsing challenges:
+### Scoring Methods
 
-- **Date chaos**: Inconsistent date formats (Jan 2020, 2020-06, 06/2021, Summer 2022)
-- **Esoteric titles**: Non-standard job titles (Chief Vibes Officer, Growth Hacker)
-- **European regional**: German-style CV with DOB, marital status, nationality
-- **Functional format**: Skills-first layout, no chronological work history
-- **Identity confusion**: Multi-part names, honorifics, maiden names, testimonial quotes
-- **Multilingual**: English with French/Spanish intertwined
-- **Prompt injection**: Hidden white text, PDF metadata injection, keyword stuffing
-- **Prose narrative**: Third-person writing, 500-word personal statement, emoji markers
-- **Same company repeat**: Same employer listed 4x with overlapping dates
-- **Skill-name collision**: Name collides with programming language (Ruby Chen at Python Solutions)
+1. **Entity-Level F1**: Entities (experience entries, education entries, etc.) are aligned between prediction and ground truth using the **Hungarian algorithm** (optimal bipartite matching via `scipy.optimize.linear_sum_assignment`). Match quality is determined by **Jaro-Winkler similarity** on key fields (threshold: 0.5).
 
-### Handwritten Resumes (28 resumes)
+2. **Description Token F1**: For matched entities with description arrays, a bag-of-words token F1 score measures description quality.
 
-Scanned handwritten resumes testing OCR and layout understanding under real-world conditions.
+3. **Hallucination Rate**: Fraction of predicted entities that have no match in ground truth (spurious predictions).
 
-### Expert-Created Templates (77 resumes)
+4. **Omission Rate**: Fraction of ground truth entities that have no match in predictions (missed extractions).
 
-Professionally designed resumes across Canva templates (63) and LaTeX templates (14) covering diverse visual layouts: single-column, two-column, sidebar, infographic, academic CV, federal resume, and more.
+All scoring is deterministic with no model in the loop.
 
-## Usage
+### Aggregate Metrics
 
-### With resume-extract-bench CLI
+Per-section scores are averaged across all resumes. The **overall F1** is the mean of all section F1 scores, providing a single leaderboard ranking metric.
 
-```bash
-pip install resume-extract-bench
-resume-bench download
-resume-bench grade-file --predictions my_predictions.jsonl --split test
-resume-bench leaderboard
-```
+---
 
-### Direct Loading
+## Leaderboard
 
-```python
-from huggingface_hub import snapshot_download
-import json
+Results from the full 155-resume benchmark (includes this dataset plus additional medium-difficulty resumes):
 
-root = snapshot_download(
-    repo_id="careerflow/ResumeExtractBench",
-    repo_type="dataset",
-)
+| Rank | Model | Overall F1 | Halluc. Rate | Omission Rate |
+|---|---|---|---|---|
+| 1 | LlamaExtract Agentic Plus | **0.945** | 5.1% | 2.9% |
+| 2 | Claude Opus | 0.937 | 5.1% | 3.6% |
+| 3 | Extend Extract | 0.937 | 4.3% | 4.3% |
+| 4 | Extend Max Context | 0.935 | 4.7% | 4.6% |
+| 5 | Reducto Extract | 0.931 | 5.4% | 4.1% |
+| 6 | Gemini 3.5 Flash | 0.928 | 5.8% | 4.3% |
+| 7 | Reducto Deep | 0.923 | 6.3% | 4.9% |
+| 8 | GPT-5.5 | 0.921 | 9.4% | 2.5% |
+| 9 | GPT-5.6 | 0.920 | 8.7% | 3.2% |
+| 10 | Gemma 4 26B | 0.914 | 6.5% | 5.4% |
 
-with open(f"{root}/test.jsonl") as f:
-    cases = [json.loads(line) for line in f if line.strip()]
+*19 models benchmarked. Full results available in the [code repository](https://github.com/careerflow/resume-extract-bench).*
 
-for case in cases:
-    pdf_path = f"{root}/{case['files']['pdf']}"
-    schema = json.load(open(f"{root}/schema.json"))
-    ground_truth = case["ground_truth"]
-    # Run your extractor on pdf_path against schema
-    # Compare output to ground_truth
-```
+---
 
-### Predictions Format
+## Tag Taxonomy
 
-Submit predictions as JSONL with `resume_id` and extracted sections:
+ResumeExtractBench tags documents along three axes:
 
-```json
-{"resume_id": "ai-engineer", "basics": {"fname": "...", "lname": "..."}, "experience": [...], ...}
-```
+| Axis | Tags | Description |
+|---|---|---|
+| **Source** | `expert-handwritten`, `synthetic-distractor` | How the resume was created |
+| **Difficulty** | `easy`, `medium`, `hard` | Overall extraction difficulty |
+| **Domain** | `engineering`, `healthcare`, `legal`, `data-science`, `government`, `general` | Professional domain |
 
-## Evaluation
+Layout tags provide additional visual/structural metadata per resume (e.g., `handwritten`, `fancy-templates`, `canva`).
 
-Scoring uses:
-- **Jaro-Winkler similarity** for field matching (threshold: 0.5)
-- **Hungarian algorithm** (scipy) for optimal entity alignment
-- **Token F1** for description quality
-- **Entity-level Precision/Recall/F1** with hallucination and omission rates
+---
 
 ## Citation
 
 ```bibtex
 @misc{careerflow2026resumeextractbench,
-  title={ResumeExtractBench: A Benchmark for Schema-Guided Resume Extraction},
-  author={Careerflow and LlamaIndex},
-  year={2026},
-  url={https://huggingface.co/datasets/careerflow/ResumeExtractBench}
+    title={ResumeExtractBench: A Benchmark for Schema-Guided Resume Extraction},
+    author={Careerflow and LlamaIndex},
+    year={2026},
+    url={https://huggingface.co/datasets/careerflow/ResumeExtractBench},
 }
 ```
 
 ## License
 
-CC-BY-4.0
+All documents are synthetic or expert-created with fictional personal information. Released under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/).
+
+## Related
+
+- **[ExtractBench](https://huggingface.co/datasets/llamaindex/ExtractBench)**: Schema-guided enterprise document extraction benchmark (370 documents, 8 business domains)
+- **[LongExtractBench-50](https://huggingface.co/datasets/micro1-inc/longextract-bench-50)**: Long-document structured extraction benchmark (50 documents)