Arcspan Cyber NER: A Multi-Source Cybersecurity Named Entity Recognition Dataset

Arcspan Cybersecurity NER Dataset is a multi-source corpus for named entity recognition in cybersecurity, covering 5 entity classes. It is built as the training and evaluation corpus for the Arcspan project, which fine-tunes OpenAI's sparse MoE Privacy Filter for cybersecurity IOC extraction. The dataset is in OPF (OpenAI Privacy Filter) JSONL format and aggregates text from threat intelligence reports, CVE descriptions, MITRE ATT&CK entries, and APT reports.

Use Cases

• Fine-tuning NER models for cybersecurity based on the 5 defined entity classes.
• Extracting Indicators of Compromise (IOCs) from threat intelligence reports mentioned in the description.
• Benchmarking model performance on security text from sources like CVE descriptions and MITRE ATT&CK entries.
• Training privacy-filtering models for security data using the OpenAI Privacy Filter format.

Strengths

• Covers 5 distinct entity classes for cybersecurity NER.
• Aggregates text from multiple specialized sources including threat intelligence reports, CVE descriptions, and MITRE ATT&CK entries.
• Formatted specifically for use with OpenAI's sparse MoE Privacy Filter architecture.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface user chairulridjal, part of the Arcspan project.

Collection Method

Aggregated from multiple cybersecurity text sources including threat intelligence reports, CVE descriptions, MITRE ATT&CK entries, and APT reports.

Freshness

Last updated 2026-05-15 11:21:05; freshness should be verified.