Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,244 @@
+---
+language:
+- en
+- zh
+- ru
+license: cc-by-4.0
+size_categories:
+- 1M<n<10M
+task_categories:
+- text-classification
+- information-retrieval
+tags:
+- cybersecurity
+- exploit
+- vulnerability
+- CVE
+- hacker-forum
+- threat-intelligence
+- NLP
+- benchmark
+pretty_name: HackerSignal
+dataset_info:
+  features:
+  - name: unified_id
+    dtype: string
+  - name: source_dataset
+    dtype: string
+  - name: source_layer
+    dtype: string
+  - name: text
+    dtype: string
+  - name: timestamp
+    dtype: string
+  - name: forum_id
+    dtype: string
+  - name: author_hash
+    dtype: string
+  - name: release_mode
+    dtype: string
+  splits:
+  - name: full
+    num_examples: 7447646
+configs:
+- config_name: default
+  data_files:
+  - split: full
+    path: unified_hacker_communities_neurips_public.jsonl
+- config_name: sample
+  data_files:
+  - split: full
+    path: hackersignal_sample_10k.jsonl
+- config_name: task1_exploit_clf
+  data_files:
+  - split: train
+    path: benchmark/task1_exploit_clf/train.jsonl
+  - split: validation
+    path: benchmark/task1_exploit_clf/val.jsonl
+  - split: test
+    path: benchmark/task1_exploit_clf/test.jsonl
+- config_name: task2_cve_linkage
+  data_files:
+  - split: train
+    path: benchmark/task2_cve_linkage/train.jsonl
+  - split: validation
+    path: benchmark/task2_cve_linkage/val.jsonl
+  - split: test
+    path: benchmark/task2_cve_linkage/test.jsonl
+  - split: corpus
+    path: benchmark/task2_cve_linkage/corpus.jsonl
+- config_name: task3_severity
+  data_files:
+  - split: train
+    path: benchmark/task3_severity/train.jsonl
+  - split: validation
+    path: benchmark/task3_severity/val.jsonl
+  - split: test
+    path: benchmark/task3_severity/test.jsonl
+- config_name: task4_hacker_exploit_labeling
+  data_files:
+  - split: train
+    path: benchmark/task4_hacker_exploit_labeling/train.jsonl
+  - split: validation
+    path: benchmark/task4_hacker_exploit_labeling/val.jsonl
+  - split: test
+    path: benchmark/task4_hacker_exploit_labeling/test.jsonl
+- config_name: task5_hacker_signal_detection
+  data_files:
+  - split: train
+    path: benchmark/task5_hacker_signal_detection/train.jsonl
+  - split: validation
+    path: benchmark/task5_hacker_signal_detection/val.jsonl
+  - split: test
+    path: benchmark/task5_hacker_signal_detection/test.jsonl
+  - split: corpus
+    path: benchmark/task5_hacker_signal_detection/corpus.jsonl
+---
+
+# HackerSignal
+
+A large-scale, multi-source dataset linking hacker community discourse, exploit databases, vulnerability advisories, and fix commits through a shared CVE identifier space.
+
+## Overview
+
+| Statistic | Value |
+|-----------|-------|
+| Documents | 7,447,646 (exact-deduplicated) |
+| Sources | 64 public forum/source identifiers |
+| Source layers | 8 |
+| Temporal span | 1988--2026 |
+| CVE-linked rows | 360,004 |
+| Benchmark tasks | 5 |
+
+## Quick Start
+
+```python
+from datasets import load_dataset
+
+# Load the 10K stratified sample (24 MB)
+sample = load_dataset("DatasetSubmission/etg-exploit-text-graph", "sample")
+
+# Load a benchmark task
+task1 = load_dataset("DatasetSubmission/etg-exploit-text-graph", "task1_exploit_clf")
+
+# Load the full corpus (12 GB)
+full = load_dataset("DatasetSubmission/etg-exploit-text-graph", "default")
+```
+
+## Source Layers
+
+| Layer | Description | Rows |
+|-------|-------------|------|
+| `hacker_community` | Forum posts from hacker/security communities | 6,974,128 |
+| `exploit_archive` | Published exploit code and advisories | 197,393 |
+| `vulnerability_reference` | NVD CVE descriptions and structured advisories | 150,916 |
+| `exploit_qa_reference` | Security Q&A and tutorial content | 68,791 |
+| `advisory_reference` | Vendor and third-party security advisories | 30,361 |
+| `bug_bounty_disclosure` | Public bug bounty reports | 13,720 |
+| `fix_commit_reference` | Vulnerability fix commit messages | 11,104 |
+| `exploitation_reference` | Active exploitation indicators (CISA KEV) | 1,233 |
+
+## Benchmark Tasks
+
+### Task 1: Exploit Relevance Classification
+Binary classification (exploit-relevant vs. not). Temporal split: train (<2022), val (2022--2023), test (2024+).
+
+| Split | Rows |
+|-------|------|
+| Train | 1,672 |
+| Val | 440 |
+| Test | 388 |
+
+### Task 2: CVE Linkage Retrieval
+Given source evidence text, retrieve the correct NVD CVE entry from a corpus of 340K descriptions. Quality-controlled to remove unretrievable rows.
+
+| Split | Rows |
+|-------|------|
+| Train | 40,698 |
+| Val | 2,551 |
+| Test | 550 |
+| Corpus | 340,536 |
+
+### Task 3: Severity Prediction
+4-class CVSS severity prediction (low/medium/high/critical) from advisory text with explicit severity strings stripped.
+
+| Split | Rows |
+|-------|------|
+| Train | 6,797 |
+| Val | 12,836 |
+| Test | 11,711 |
+
+### Task 4: Hacker Exploit Labeling
+Ternary weak-supervision task: 0=non-exploit/noise, 1=vulnerability discussion, 2=actionable exploit.
+
+| Split | Rows |
+|-------|------|
+| Train | 50,382 |
+| Val | 1,933 |
+| Test | 25,963 |
+
+### Task 5: Hacker Exploit Signal Detection
+Joint detection + retrieval: identify actionable exploit signals and link to NVD CVE context.
+
+| Split | Rows |
+|-------|------|
+| Train | 1,146 |
+| Val | 438 |
+| Test | 111 |
+| Corpus | 340,536 |
+
+## Schema
+
+Each record in the main corpus contains:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `unified_id` | string | SHA-256-derived release identifier |
+| `source_dataset` | string | Input source file or dataset name |
+| `source_layer` | string | Normalized layer (see above) |
+| `text` | string | UTF-8 post/advisory text (max 8000 chars) |
+| `timestamp` | string | ISO 8601 publication datetime (UTC) |
+| `forum_id` | string | Source identifier |
+| `author_hash` | string | SHA-256 pseudonymized author |
+| `release_mode` | string | Governance mode (see below) |
+
+## Release Governance
+
+Records follow source-specific release modes:
+
+- **`redistributable_text`** (5 sources): Full text included under public-domain or CC licenses.
+- **`research_text_with_terms`** (9 sources): Text included with source attribution; no commercial reuse.
+- **`metadata_or_pointer_only`** (13 sources): Only metadata, timestamps, CVE refs, text hashes, and lengths; raw text withheld.
+
+See `docs/release_governance.md` for the full source matrix, takedown policy, and responsible-use restrictions.
+
+## Responsible Use
+
+This dataset is intended for **defensive cybersecurity research** only. Prohibited uses include:
+- Training systems for automated exploit code generation or malware
+- De-anonymizing forum participants outside approved research protocols
+- Operational blocking or law-enforcement decisions based solely on model outputs
+- Republishing text from `metadata_or_pointer_only` sources
+
+## Limitations
+
+- English-language bias; non-English communities may be underrepresented
+- Historical forum imports skew toward 2015--2017
+- Task labels are weak (CVE-linkage-derived), not human-annotated
+- ~8% near-duplicate content remains after exact deduplication
+- 1.57M rows under 8 tokens (useful for interaction analysis, not classification)
+
+## Citation
+
+```bibtex
+@inproceedings{hackersignal2026,
+  title={HackerSignal: A Multi-Source Dataset for Hacker Community Analysis and CVE Linkage},
+  author={Anonymous},
+  booktitle={NeurIPS Evaluations \& Datasets Track},
+  year={2026}
+}
+```
+
+## License
+
+Code and metadata: CC BY 4.0. Source text retains source-specific terms documented in the governance addendum and release manifest.