{
  "researched_on": "2026-11-04",
  "purpose": "Evidence for bilingual company mention to listed issuer linking; no model quality result is claimed.",
  "benchmarks": [
    {
      "name": "FinNewsBench",
      "language": "ko",
      "source": "https://huggingface.co/datasets/davidkim205/FinNewsBench",
      "verified_local_rows": 310,
      "tasks": [
        "company extraction",
        "person extraction",
        "themes",
        "keywords",
        "central company sentiment"
      ],
      "provenance": "Synthetic articles; LLM annotations with expert verification according to publisher card.",
      "license_declared": "CC-BY-NC-4.0",
      "usable_for": "Research evaluation of structured extraction and company/person separation.",
      "gaps": [
        "No ticker gold field",
        "Synthetic rather than independent real news",
        "Small sample",
        "Noncommercial license; not assumed cleared for commercial training"
      ],
      "access_verified": "Downloaded dataset.jsonl and parsed 310 rows; author linked GitHub repository returned 404."
    },
    {
      "name": "SEntFiN 1.0",
      "language": "en",
      "source": "https://arxiv.org/abs/2305.12257",
      "data_source": "https://github.com/pyRis/SEntFiN",
      "verified_local_rows": 10753,
      "verified_local_entity_rows": 1010,
      "tasks": [
        "entity-specific sentiment in headlines"
      ],
      "provenance": "Human-annotated headlines; paper reports 2847 multi-entity headlines.",
      "license_declared": "Repository LICENSE is MIT",
      "usable_for": "Multi-entity evaluation design and sentiment baseline; inspect individual source rights before redistribution.",
      "gaps": [
        "Predominantly Indian finance names and symbols",
        "Alias table also includes sector labels",
        "Reported sentiment accuracy is not end-to-end ticker precision"
      ],
      "access_verified": "Downloaded and parsed both news CSV and entity alias CSV."
    },
    {
      "name": "KLUE NER",
      "language": "ko",
      "source": "https://github.com/KLUE-benchmark/KLUE",
      "tasks": [
        "general named entity recognition"
      ],
      "license_declared": "CC-BY-SA-4.0",
      "usable_for": "Korean mention boundaries and entity type separation.",
      "gaps": [
        "General NER, not a financial issuer/ticker linking benchmark"
      ],
      "access_verified": "Official repository and README checked; dataset not executed."
    },
    {
      "name": "BloombergGPT NER+NED",
      "language": "en",
      "source": "https://arxiv.org/html/2303.17564v3",
      "tasks": [
        "company mentions to ticker sets"
      ],
      "usable_for": "Task definition and separation of NER from issuer linking.",
      "gaps": [
        "Internal Bloomberg documents",
        "Ticker gold generated with an existing linker",
        "Not a downloadable independent public benchmark"
      ],
      "access_verified": "Paper section Exploratory Task: NER+NED checked."
    },
    {
      "name": "FinEntity",
      "language": "en",
      "source": "https://arxiv.org/abs/2310.12406",
      "data_source": "https://huggingface.co/datasets/yixuantt/FinEntity",
      "verified_local_rows": 979,
      "verified_annotations": 2131,
      "tasks": [
        "financial entity spans and sentiment"
      ],
      "license_declared": "ODC-BY on publisher HF card",
      "gaps": [
        "Not ticker gold",
        "Includes commodities/sectors/brands",
        "70 annotations mismatch raw text offsets",
        "Old public news; model pretraining contamination unknown"
      ],
      "access_verified": "Downloaded original JSON; no offset repairs applied; clean-subset review queue frozen separately"
    },
    {
      "name": "FNSPID",
      "language": "Primarily US stock news; Korean issuer coverage not established",
      "source": "https://arxiv.org/html/2402.06698v1",
      "data_source": "https://github.com/Zdong104/FNSPID_Financial_News_Dataset",
      "tasks": ["News and stock-price time-series pairing"],
      "publisher_reported_news_records": 15698563,
      "verified_local_rows": 0,
      "usable_for": "Potential large external stress-test corpus; individual source/symbol pairs require validation before becoming reference labels",
      "gaps": [
        "A symbol column does not establish exhaustive independently reviewed news-to-issuer gold",
        "Old public news does not establish unseen-current-news accuracy",
        "Repository README contains both a July 2025 rights-release update and an older commercial-use prohibition; permission scope not resolved here"
      ],
      "access_verified": "Primary paper and publisher repository inspected; no bulk download, code execution, or deployment"
    },
    {
      "name": "FinTexTS",
      "language": "US-company English-language examples; Korean performance not established",
      "source": "https://arxiv.org/html/2603.02702v1",
      "data_source": "https://huggingface.co/datasets/EXAONE-BI/FinTexTS",
      "code_source": "https://github.com/leejaehoon2016/FinTexTS",
      "model_source": "https://huggingface.co/EXAONE-BI/FinTexTS-Embedding",
      "tasks": ["Macro, sector, target-company and related-company news pairing for forecasting"],
      "publisher_reported_companies": 100,
      "verified_local_rows": 0,
      "usable_for": "Architecture reference for a future separately evaluated relevance layer: filing profile, semantic retrieval, contextual classification and irrelevant filtering",
      "gaps": [
        "Forecasting utility is not news-to-current-ticker identity precision or recall",
        "Published dataset viewer contains dated ticker summaries and filing fields rather than an exhaustive independently adjudicated raw-article mention-to-issuer test set",
        "Target-company influence and related-company influence must not be used as direct-mention identity labels",
        "Korean language validation and model/data reuse permissions not established"
      ],
      "embedding_details": {
        "publisher_base": "theme-emb/linq-v2",
        "architecture": "MistralModel plus mean pooling",
        "dimensions": 4096,
        "evaluated_locally": false
      },
      "access_verified": "Primary paper, publisher dataset viewer, model card and actual repository scripts list inspected; not installed or executed"
    }
  ],
  "implementation_references": [
    {
      "name": "JEL",
      "source": "https://arxiv.org/html/2411.02695v1",
      "lesson": "Company linking to an enterprise knowledge graph uses context and entity representations; internal results do not transfer to this project automatically."
    },
    {
      "name": "GLiNER multilingual v2.1",
      "source": "https://huggingface.co/urchade/gliner_multi-v2.1",
      "license_declared": "Apache-2.0 for this specific checkpoint",
      "role": "Candidate company/person mention extraction; Korean and English quality must be measured locally. Does not provide issuer linking.",
      "execution_status": "Evaluated on iMac in job 20261104T124834Z-94d107a33af7; exact mention coverage remains a component diagnostic, not ticker identity quality. See implementation_status.json."
    },
    {
      "name": "BLINK",
      "source": "https://github.com/facebookresearch/BLINK",
      "license_declared": "MIT repository",
      "role": "Reference architecture: candidate retrieval followed by context/entity reranking. Wikipedia checkpoint is not a current KRX/US issuer master.",
      "execution_status": "not installed or evaluated"
    }
  ],
  "conclusion": "Public resources support component evaluation. No reviewed source supplies a ready-made bilingual real-news-to-current-KRX-and-US-ticker benchmark. An independent project test set is required."
}