File size: 20,933 Bytes
3f64c20 dac8248 062f4f3 3f64c20 496b61e 3f64c20 496b61e 3f64c20 496b61e 3f64c20 b0bfbd1 3f64c20 496b61e 3f64c20 062f4f3 3f64c20 062f4f3 3f64c20 062f4f3 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 dac8248 3f64c20 | 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 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 | {
"@context": {
"@language": "en",
"@vocab": "https://schema.org/",
"cr": "http://mlcommons.org/croissant/",
"sc": "https://schema.org/",
"dct": "http://purl.org/dc/terms/",
"rai": "http://mlcommons.org/croissant/RAI/",
"prov": "http://www.w3.org/ns/prov#"
},
"@type": "sc:Dataset",
"dct:conformsTo": "http://mlcommons.org/croissant/RAI/1.0",
"name": "HackerSignal",
"description": "HackerSignal is a large-scale, multi-source dataset linking hacker community discourse, exploit databases, vulnerability advisories, and fix commits through a shared CVE identifier space. It spans 64 public forum/source identifiers across eight source layers, covers 1988\u20132026, and supports three benchmark tasks: CVE linkage retrieval, hacker signal detection with CVE grounding, and CVE-disjoint temporal generalization.",
"url": "https://huggingface.co/datasets/DatasetSubmission/HackerSignal",
"version": "1.0.0",
"license": "https://creativecommons.org/licenses/by/4.0/",
"usageInfo": "Code and metadata are CC BY 4.0 unless otherwise noted. Source text follows source-specific release modes documented in docs/release_governance.md; terms-ambiguous sources may be metadata_or_pointer_only.",
"isAccessibleForFree": true,
"datePublished": "2026",
"inLanguage": [
"en",
"zh",
"ru"
],
"citation": "Anonymous. (2026). HackerSignal \u2014 A Multi-Source Dataset for Hacker Community Analysis and CVE Linkage. Proceedings of NeurIPS Evaluations & Datasets Track.",
"creator": [
{
"@type": "sc:Person",
"name": "Anonymous",
"email": "anonymous@example.com",
"affiliation": "Anonymous"
}
],
"keywords": [
"cybersecurity",
"exploit",
"vulnerability",
"CVE",
"hacker forum",
"threat intelligence",
"NLP",
"information retrieval",
"text classification",
"benchmark"
],
"prov:wasDerivedFrom": [
{
"@type": "cr:FileObject",
"name": "NVD (National Vulnerability Database)",
"contentUrl": "https://services.nvd.nist.gov/rest/json/cves/2.0",
"description": "US government public-domain CVE records via REST API."
},
{
"@type": "cr:FileObject",
"name": "CISA KEV (Known Exploited Vulnerabilities)",
"contentUrl": "https://www.cisa.gov/known-exploited-vulnerabilities-catalog",
"description": "US government public-domain catalog of actively exploited vulnerabilities."
},
{
"@type": "cr:FileObject",
"name": "GitHub Advisory Database",
"contentUrl": "https://github.com/advisories",
"description": "Security advisories via GraphQL API, licensed CC BY 4.0."
},
{
"@type": "cr:FileObject",
"name": "ExploitDB",
"contentUrl": "https://www.exploit-db.com/",
"description": "Public exploit archive, licensed CC BY-SA 4.0."
},
{
"@type": "cr:FileObject",
"name": "CVEfixes",
"contentUrl": "https://zenodo.org/records/7029359",
"description": "Fix commits and CVE metadata from Zenodo, licensed CC BY 4.0."
},
{
"@type": "cr:FileObject",
"name": "HackerOne Public Reports",
"contentUrl": "https://huggingface.co/datasets",
"description": "Public bug bounty reports from HuggingFace Hub."
},
{
"@type": "cr:FileObject",
"name": "DeepDarkCTI",
"contentUrl": "https://github.com/fastfire/deepdarkCTI",
"description": "Cyber threat intelligence from GitHub public dataset."
}
],
"prov:wasGeneratedBy": {
"@type": "prov:Activity",
"name": "HackerSignal ETL Pipeline",
"description": "Collection scripts executed between January and April 2026 using public APIs, public dataset ingestion, web scraping of publicly accessible pages, and pre-processed dataset imports. The pipeline performs HTML stripping, encoding normalization, text truncation (8000 chars), exact deduplication by SHA-256 fingerprint, author pseudonymization, and source-layer normalization.",
"prov:startedAtTime": "2026-01-01",
"prov:endedAtTime": "2026-04-30"
},
"rai:dataCollection": "Data was collected through four mechanisms: (1) Public APIs \u2014 NVD (NIST REST), GitHub Advisory Database (GraphQL), CISA KEV (JSON feed), HackerOne (HuggingFace Hub); (2) Public dataset ingestion \u2014 Gayanku (Kaggle), HackForums (Kaggle), DTL Exploits (GitHub/AZSecure), CVEfixes (Zenodo), DeepDarkCTI (GitHub), Kaeli hacker forums (Kaggle), CrackingArena (Kaggle), ZeroDay (GitHub); (3) Web scraping and public forum APIs \u2014 ExploitDB (CSV), Vulnerability-Lab, ZeroScience Lab, 0x00sec, HackerSploit, ParrotSec, Hack The Box, Go4Expert, Full Disclosure mailing list archive, Seebug; (4) Pre-processed datasets from HuggingFace Hub. All sources were publicly accessible at collection time; no authentication was bypassed.",
"rai:dataCollectionType": "Web scraping, API, Surveys/Interviews",
"rai:dataCollectionTimeframe": {
"@type": "sc:DateTime",
"startDate": "2026-01-01",
"endDate": "2026-04-30"
},
"rai:dataCollectionRawData": "Raw scraped HTML is not distributed. The processed JSONL files are the primary artifact. Collection and preprocessing scripts are available in the src/etg/collectors/ directory of the code repository.",
"rai:dataCollectionMissingData": "Coverage is limited by public API availability, rate limits, and pagination depth for forum sources. Posts behind authentication walls are excluded. Some timestamps reflect scraping date rather than original publication date. Author information is pseudonymized and not recoverable. Some legacy forum imports do not preserve thread structure.",
"rai:dataPreprocessingProtocol": "HTML tags stripped from forum post content; HTML entities unescaped; text truncated at 8000 characters; encoding errors replaced with Unicode replacement character (U+FFFD); source provenance normalized into source_dataset, source_layer, forum_id, and optional thread/category fields; exact duplicates removed across sources by normalized-text SHA-256 fingerprint; Task 3 benchmark inputs strip explicit severity labels, CVSS score lines, and CVSS vectors after label extraction; author strings pseudonymized via SHA-256 hashing of source-namespaced author identifiers.",
"rai:dataLimitations": "Forum sources skew toward English-language Western cybersecurity communities; threats emerging from non-English communities may be underrepresented. High-volume historical forum imports skew toward 2015\u20132017 or earlier. Task 1 labels are weak CVE-linkage labels, not human exploit-semantics annotations; CVE-index absence is not a manual negative annotation. Task 2 signal detection uses weak labels derived from CVE metadata and actionability cues, not human annotations. Task 3 shares the same label methodology as Task 1 under a stricter CVE-disjoint constraint. Approximately 8.18% near-duplicate content remains after exact deduplication. The corpus contains 1,569,263 rows under eight tokens useful for interaction analysis but not text classification.",
"rai:dataBiases": "English-language bias: the corpus is predominantly English with multilingual content (Russian, Turkish, Persian, Arabic, Chinese) limited to deepdarkcti and several community sources. Temporal bias: bulk of structured NVD records begin in 2002; high-volume forum imports cluster around 2015\u20132017. Source selection bias: only publicly indexed or API-accessible sources are included, potentially underrepresenting closed or invitation-only communities. CISA KEV covers only actively exploited vulnerabilities, over-representing critical severity relative to true distribution. Western-centric community selection may underrepresent threat actors operating primarily in non-English forums.",
"rai:personalSensitiveInformation": "No private communications, medical records, financial records, or government classified information is intentionally included. All text was collected from publicly accessible sources. Author strings are pseudonymized via SHA-256 hashing of source-namespaced author identifiers; original author strings are not recoverable from the released dataset. Public forum usernames are not distributed. The release supports takedown and correction requests through the repository issue tracker.",
"rai:dataUseCases": "Intended uses: defensive cybersecurity research, temporal retrieval, provenance-aware benchmark design, vulnerability lifecycle analysis, exploit prediction and early warning systems, cross-lingual threat intelligence, graph neural network research on exploit\u2013CVE\u2013fix commit knowledge graphs, and pre-training domain-adapted language models for cybersecurity. Prohibited uses: training systems for automated exploit code generation or malware; de-anonymizing forum participants or linking pseudonymous identities outside approved research protocols; operational blocking, law-enforcement, or employment decisions based solely on model outputs; republishing source text from metadata_or_pointer_only sources.",
"rai:dataSocialImpact": "HackerSignal enables earlier detection of vulnerability exploitation activity by connecting hacker community discourse to the CVE lifecycle. Positive impacts include faster defensive response times and better understanding of exploit emergence patterns. Potential negative impacts include misuse for automated attack generation or targeting of individuals; these are mitigated through responsible-use restrictions, author pseudonymization, source-level release governance, and explicit prohibition of offensive applications in the dataset license.",
"rai:dataReleaseMaintenancePlan": "Maintained by the dataset authors. Annual updates planned to incorporate new CVE records, CISA KEV entries, and newly published advisories. Versioning follows semantic versioning (MAJOR.MINOR.PATCH). All tagged versions remain available on HuggingFace Hub. Takedown and correction requests acknowledged within 10 business days; disputed text removed in next patch release. Sources that change terms post-release are downgraded to metadata_or_pointer_only until permissions are clarified.",
"rai:hasSyntheticData": "none",
"distribution": [
{
"@type": "cr:FileObject",
"@id": "hf-repo",
"name": "HuggingFace Repository",
"contentUrl": "https://huggingface.co/datasets/DatasetSubmission/HackerSignal",
"encodingFormat": "application/jsonlines",
"sha256": "See versioned manifest checksums in the HuggingFace repository."
},
{
"@type": "cr:FileObject",
"@id": "public-jsonl",
"name": "unified_hacker_communities_neurips_public.jsonl",
"description": "Governed public release file enforcing source-level release modes from data/release/source_release_manifest.json.",
"contentUrl": "https://huggingface.co/datasets/DatasetSubmission/HackerSignal/resolve/main/data/unified_hacker_communities_neurips_public.jsonl",
"encodingFormat": "application/jsonlines"
},
{
"@type": "cr:FileObject",
"@id": "source-manifest",
"name": "source_release_manifest.json",
"description": "Machine-readable source provenance and release-mode manifest with checksums and row counts.",
"contentUrl": "https://huggingface.co/datasets/DatasetSubmission/HackerSignal/resolve/main/data/release/source_release_manifest.json",
"encodingFormat": "application/json"
}
],
"recordSet": [
{
"@type": "cr:RecordSet",
"@id": "posts",
"name": "posts",
"description": "Forum posts, exploit advisories, vulnerability descriptions, and fix-commit messages from 64 public forum/source identifiers. Each record is one document. 7,447,646 exact-deduplicated rows spanning 1988\u20132026.",
"field": [
{
"@type": "cr:Field",
"@id": "posts/unified_id",
"name": "unified_id",
"description": "SHA-256-derived release identifier.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/source_dataset",
"name": "source_dataset",
"description": "Input source file or dataset name.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/source_layer",
"name": "source_layer",
"description": "Normalized layer: hacker_community, exploit_archive, vulnerability_reference, fix_commit_reference, exploit_qa_reference, or exploitation_reference.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/text",
"name": "text",
"description": "Full text of the post or advisory (UTF-8, max 8000 chars). Withheld for metadata_or_pointer_only sources.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/text_raw",
"name": "text_raw",
"description": "Source text before release cleaning, where available. Withheld for metadata_or_pointer_only sources.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/timestamp",
"name": "timestamp",
"description": "Publication date-time (ISO 8601, UTC).",
"dataType": "sc:DateTime"
},
{
"@type": "cr:Field",
"@id": "posts/forum_id",
"name": "forum_id",
"description": "Source identifier (e.g. 'exploitdb', 'nvd', 'hackforums').",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/author_hash",
"name": "author_hash",
"description": "SHA-256 hash of the source-namespaced author string (pseudonymised).",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/release_mode",
"name": "release_mode",
"description": "Release-governance mode: redistributable_text, research_text_with_terms, metadata_or_pointer_only, or excluded.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/thread_url",
"name": "thread_url",
"description": "Source thread URL when exposed by the source.",
"dataType": "sc:URL"
},
{
"@type": "cr:Field",
"@id": "posts/thread_title",
"name": "thread_title",
"description": "Thread or advisory title when available.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "posts/section",
"name": "section",
"description": "Forum section or category when available.",
"dataType": "sc:Text"
}
]
},
{
"@type": "cr:RecordSet",
"@id": "cve_index",
"name": "cve_index",
"description": "Cross-source CVE linkage index mapping post IDs to CVE identifiers. One row per (post, CVE) pair. 360,004 rows contain CVE identifiers.",
"field": [
{
"@type": "cr:Field",
"@id": "cve_index/exploit_id",
"name": "exploit_id",
"description": "Foreign key to posts/unified_id.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "cve_index/cve_id",
"name": "cve_id",
"description": "CVE identifier in the form CVE-YYYY-NNNNN.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "cve_index/exploit_text",
"name": "exploit_text",
"description": "First 500 characters of the source post text.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "cve_index/title",
"name": "title",
"description": "Advisory or post title.",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "cve_index/published",
"name": "published",
"description": "Publication date-time (ISO 8601, UTC).",
"dataType": "sc:DateTime"
},
{
"@type": "cr:Field",
"@id": "cve_index/source",
"name": "source",
"description": "Source identifier matching posts/forum_id.",
"dataType": "sc:Text"
}
]
},
{
"@type": "cr:RecordSet",
"@id": "task1_cve_linkage",
"name": "task1_cve_linkage",
"description": "Task 1 \u2014 CVE Linkage Retrieval (CVE-R). Cross-source temporally OOD entity grounding: given exploit/advisory evidence text, retrieve the correct NVD CVE entry from a corpus of 340,536 descriptions. Quality-controlled to remove unretrievable rows. 40,698 train / 2,551 val / 550 test.",
"field": [
{
"@type": "cr:Field",
"@id": "task1/id",
"name": "id",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task1/text",
"name": "text",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task1/cve_id",
"name": "cve_id",
"dataType": "sc:Text",
"description": "Target CVE identifier in the corpus."
},
{
"@type": "cr:Field",
"@id": "task1/source",
"name": "source",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task1/timestamp",
"name": "timestamp",
"dataType": "sc:DateTime"
},
{
"@type": "cr:Field",
"@id": "task1/split",
"name": "split",
"dataType": "sc:Text",
"description": "train / val / test"
}
]
},
{
"@type": "cr:RecordSet",
"@id": "task2_signal_detection",
"name": "task2_signal_detection",
"description": "Task 2 \u2014 Hacker Signal Detection (HSD). Two-stage: detect actionable exploit signals from hacker community noise (10:1 negative ratio), then ground positives to NVD CVE. 4,202 train / 1,606 val / 407 test.",
"field": [
{
"@type": "cr:Field",
"@id": "task2/id",
"name": "id",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task2/text",
"name": "text",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task2/actionable_label",
"name": "actionable_label",
"dataType": "sc:Integer",
"description": "0=not_actionable, 1=actionable_exploit_signal"
},
{
"@type": "cr:Field",
"@id": "task2/cve_id",
"name": "cve_id",
"dataType": "sc:Text",
"description": "Primary CVE target for actionable records; empty for non-actionable."
},
{
"@type": "cr:Field",
"@id": "task2/source",
"name": "source",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task2/timestamp",
"name": "timestamp",
"dataType": "sc:DateTime"
},
{
"@type": "cr:Field",
"@id": "task2/split",
"name": "split",
"dataType": "sc:Text",
"description": "train / val / test"
}
]
},
{
"@type": "cr:RecordSet",
"@id": "task3_temporal_generalization",
"name": "task3_temporal_generalization",
"description": "Task 3 \u2014 Temporal Generalization (TG). Identical retrieval formulation to Task 1 but with strict CVE-disjoint constraint: C_train \u2229 C_test = \u2205. Split by CVE publication year (train <2022, test 2024+). 41,630 train / 1,776 val / 393 test.",
"field": [
{
"@type": "cr:Field",
"@id": "task3/id",
"name": "id",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task3/text",
"name": "text",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task3/cve_id",
"name": "cve_id",
"dataType": "sc:Text",
"description": "Target CVE identifier (from 2024+ CVE set, disjoint from training CVEs)."
},
{
"@type": "cr:Field",
"@id": "task3/source",
"name": "source",
"dataType": "sc:Text"
},
{
"@type": "cr:Field",
"@id": "task3/timestamp",
"name": "timestamp",
"dataType": "sc:DateTime"
},
{
"@type": "cr:Field",
"@id": "task3/split",
"name": "split",
"dataType": "sc:Text",
"description": "train / val / test"
}
]
}
]
} |