Update README.md

README.md

@@ -6,7 +6,8 @@ tags:
 - cybersecurity
 - mitre-attack
 - ioc-extraction
-- anomaly-detection
+- phishing-detection
+- log-analysis
 pretty_name: Astraea Unified Threat Dataset v4
 size_categories: 100K<n<1M
 task_categories:
@@ -16,50 +17,142 @@ task_categories:
 
 # Astraea Unified Threat Dataset (v4)
 
-Developed by **Auren Research**, **Astraea Unified Threat (v4)** is a highly optimized, production-balanced telemetry dataset engineered for fine-tuning modern multi-task encoder architectures (such as `chandar-lab/NeoBERT` and `answerdotai/ModernBERT`). 
+Developed by **Auren Research**, **Astraea Unified Threat (v4)** is a production-balanced telemetry dataset for fine-tuning multi-task encoder architectures such as [`answerdotai/ModernBERT`](https://huggingface.co/answerdotai/ModernBERT-base) and [`chandar-lab/NeoBERT`](https://huggingface.co/chandar-lab/NeoBERT).
 
-By aggregating raw event streams into chronological multi-line sessions, Astraea enables encoder models to learn long-context temporal correlations, compute continuous anomaly scores, and perform token-level named entity recognition (NER) on indicators of compromise.
+Each example is a structured security log or email payload with task-specific prefixes (for example, `[TASK=LOG_CLS]`), enabling a single model to perform log triage, IOC extraction, MITRE ATT&CK mapping, attack-chain classification, and phishing detection.
+
+---
+
+## Files
+
+| File | Rows | Size | Purpose |
+|---|---:|---:|---|
+| `train.parquet` | 220,295 | ~84 MB | Primary training split |
+| `val.parquet` | 31,843 | ~14 MB | Validation for log classification and IOC NER |
+| `test.parquet` | 20,000 | ~22 MB | Held-out public phishing benchmark |
+| `astraea-unified-threat.parquet` | 270,263 | ~127 MB | Legacy unified export (optional) |
+
+**Recommended for training:** use `train.parquet`, `val.parquet`, and `test.parquet`. The unified file does not include all rows from the split release (2,971 additional training examples and a task-aware partition are only in the split files).
 
 ---
 
 ## Dataset Profile
 
-| Metric | Value | Status / Notes |
+| Metric | Value | Notes |
 |---|---|---|
-| **Total Rows** | 274,778 | Chronological sessionized log streams |
-| **Class Balance (`is_malicious`)** | **44.7%** | Mathematically balanced to mirror real-world production |
-| **Mean Anomaly Score** | 0.458 | Ground-truth continuous target variable [0.0, 1.0] |
-| **Validated IOC Spans** | 1,136,768 | Character offsets programmatically verified |
-| **MITRE Coverage** | 822 unique techniques | Documented across 30 primary and sub-tactics |
-| **Language** | 100% English | Standard telemetry syntax and markdown Q&A |
-| **Format** | Parquet (zstd) | 132.0 MB compressed footprint |
+| **Total rows (splits)** | 272,138 | train + val + test |
+| **Class balance (`is_malicious`)** | **44.2% malicious** | Globally balanced across splits |
+| **Mean anomaly score** | 0.443 | Continuous target in [0.0, 1.0] |
+| **Validated IOC spans** | 1,134,937 | Character offsets verified on IOC rows |
+| **MITRE coverage** | 19 techniques, 10 tactics | On `mitre_classification` rows |
+| **Log types** | 21 | Windows, Linux, cloud, EDR, network, IAM, K8s, WAF, email, and more |
+| **Sources** | 23 | Synthetic telemetry + public phishing |
+| **Language** | English | Standard telemetry and email syntax |
+| **Format** | Parquet | Columnar, Hugging Face–compatible |
+
+---
+
+## Train / Validation / Test Splits
+
+Splits are **task-aware** and **leakage-safe**: no overlapping `text` values across train, val, or test.
+
+### `train.parquet` (220,295 rows)
+
+| Task prefix | `label_type` | Rows |
+|---|---|---:|
+| `[TASK=LOG_CLS]` | `log_classification` | 126,805 |
+| `[TASK=IOC_NER]` | `ioc_extraction` | 48,231 |
+| `[TASK=MITRE]` | `mitre_classification` | 24,650 |
+| `[TASK=ATTACK_CHAIN]` | `attack_chain` | 12,000 |
+| `[TASK=PHISH]` | `phishing_detection` | 8,609 |
+
+Use for fine-tuning all supported tasks. Includes synthetic phishing (`synthetic_email`) plus adversarial and false-positive hard negatives.
+
+### `val.parquet` (31,843 rows)
+
+| Task prefix | `label_type` | Rows |
+|---|---|---:|
+| `[TASK=LOG_CLS]` | `log_classification` | 24,405 |
+| `[TASK=IOC_NER]` | `ioc_extraction` | 7,438 |
+
+Focused on **network and supply-chain telemetry** (netflow, CI/CD, HTTP, firewall, DNS). Malicious rate for log classification is **~19.1%** (vs. ~4.3% in train), providing a harder validation signal than random sampling.
+
+### `test.parquet` (20,000 rows)
+
+| Task prefix | `label_type` | Rows |
+|---|---|---:|
+| `[TASK=PHISH]` | `phishing_detection` | 20,000 |
+
+All rows are from **`public_phishing`** (Seven Phishing Email Datasets). Use this split for real-world phishing evaluation only.
+
+> **Note:** MITRE and attack-chain tasks currently have no dedicated test split. Hold out a subset of `train.parquet` or report metrics on validation data for those tasks.
+
+---
+
+## Supported Tasks
+
+| Task | Head type | Label columns | Eval split |
+|---|---|---|---|
+| Log triage | Sequence classification | `is_malicious` | `val` |
+| IOC extraction | Token classification (NER) | `ioc_spans` | `val` |
+| MITRE mapping | Multi-label classification | `mitre_tactic`, `mitre_technique` | train holdout |
+| Attack chain | Sequence / multi-label classification | `attack_stage` | train holdout |
+| Phishing detection | Sequence classification | `is_malicious` | `test` |
 
 ---
 
 ## Technical Features
 
-* **2026 Threat Vector Telemetry:** Incorporates high-fidelity simulated telemetry covering modern enterprise attack surfaces: Non-Human Identity (NHI) and Service Account token hijacking, CI/CD pipeline/OAuth integration compromises, and highly contextualized generative AI spearphishing artifacts.
-* **Benign Background Baseline:** Includes 140,000 baseline corporate sessions (Windows Event Viewer, Linux Syslog, Database transactional logs, read-only CloudTrail/Azure Activity audits, and EDR heartbeat telemetry) to train models against false positive saturation.
-* **Token Boundary Fuzzing:** All network addresses (IPs), domains, cryptographic hashes, and OS filenames are systematically randomized across splits to enforce behavioral pattern learning over static string memorization.
+- **2026 threat-vector telemetry:** Simulated enterprise attack surfaces including NHI/service-account abuse, CI/CD and OAuth compromise patterns, cloud audit anomalies, EDR alerts, and generative-AI-style spearphishing artifacts.
+- **Benign background baseline:** Large volumes of benign Windows, Linux, database, cloud, and EDR telemetry to reduce false-positive saturation in production-like settings.
+- **Malicious augmentations:** Additional malicious Windows, Linux, EDR, cloud, and netflow examples (not present in the legacy unified export) so models cannot rely on log-type shortcuts.
+- **Token boundary fuzzing:** IPs, domains, hashes, and filenames are randomized across examples to encourage behavioral pattern learning over memorization.
+- **Public holdout:** All 20,000 public phishing emails are isolated in `test.parquet` for unbiased evaluation.
 
 ---
 
 ## Data Schema
 
-Each row contains 12 structured columns for unified multi-head architecture training:
-
-* `row_id` *(int64)*: Shuffled unique row identifier.
-* `text` *(string)*: Raw log stream chunk or email payload. **Primary model feature input.**
-* `source` *(string)*: Dataset origin label (e.g., `synthetic_cloud`, `public_phishing`).
-* `label_type` *(string)*: Downstream classification head director (`ioc_extraction`, `mitre_classification`, etc.).
-* `mitre_tactic` *(string)*: MITRE ATT&CK core tactic or `none`.
-* `mitre_technique` *(string)*: MITRE identifier (e.g., `T1059.001`) or `none`.
-* `ioc_spans` *(JSON string)*: Span alignment maps: `[{"start": int, "end": int, "type": str, "value": str}]`.
-* `anomaly_score` *(float32)*: Continuous target density metric [0.0, 1.0].
-* `attack_stage` *(string)*: Cyber Kill Chain sequence categorization.
-* `is_malicious` *(bool)*: Binary threat classification ground truth.
-* `log_type` *(string)*: Telemetry sub-type classifier (e.g., `windows_event`, `kubernetes`, `iam`, `waf`).
-* `metadata` *(JSON string)*: Ancillary technical provenance properties.
+Each row contains 12 structured columns:
+
+| Column | Type | Description |
+|---|---|---|
+| `row_id` | int64 | Row index **within each split file** (not a global ID across files) |
+| `text` | string | Log stream or email payload; primary model input |
+| `source` | string | Provenance label (e.g. `synthetic_cloud`, `public_phishing`) |
+| `label_type` | string | Task router (`log_classification`, `ioc_extraction`, etc.) |
+| `mitre_tactic` | string | MITRE ATT&CK tactic or `none` |
+| `mitre_technique` | string | MITRE technique ID (e.g. `T1059.001`) or `none` |
+| `ioc_spans` | JSON string | Span annotations: `[{"start": int, "end": int, "type": str, "value": str}]` |
+| `anomaly_score` | float | Continuous score in [0.0, 1.0] |
+| `attack_stage` | string | Kill-chain stage label(s) |
+| `is_malicious` | bool | Binary threat label |
+| `log_type` | string | Telemetry subtype (e.g. `windows_event`, `kubernetes`, `iam`) |
+| `metadata` | JSON string | Ancillary provenance fields |
+
+### IOC span types
+
+`ip`, `domain`, `email`, `filename`, `hash_md5`, `hash_sha256`, `hash_sha1`, `punycode`, `registry_key`
+
+---
+
+## Quick Start
+
+```python
+from datasets import load_dataset
+
+# Load from a local directory or Hugging Face Hub
+train = load_dataset("parquet", data_files="train.parquet", split="train")
+val   = load_dataset("parquet", data_files="val.parquet", split="train")
+test  = load_dataset("parquet", data_files="test.parquet", split="train")
+
+# Example: log classification
+logs = train.filter(lambda x: x["label_type"] == "log_classification")
+print(logs[0]["text"])
+print(logs[0]["is_malicious"])
+```
+
+For multi-task fine-tuning, pass the full `text` field (including the `[TASK=...]` prefix) to models such as ModernBERT, and use task-specific classification or token heads keyed on `label_type`.
 
 ---
 
@@ -71,5 +164,6 @@ Each row contains 12 structured columns for unified multi-head architecture trai
   author       = {Auren Research},
   year         = {2026},
   publisher    = {Hugging Face},
-  howpublished = {\url{[https://huggingface.co/auren-research/astraea-unified-threat](https://huggingface.co/auren-research/astraea-unified-threat)}}
-}
+  howpublished = {\url{https://huggingface.co/auren-research/astraea-unified-threat}}
+}
+```