Update README.md

README.md

@@ -1,120 +1,18 @@
----
-license: apache-2.0
-task_categories:
-- text-classification
-- other
-tags:
-- logs
-- cybersecurity
-- attack-detection
-- network-security
-size_categories:
-- 100K<n<1M
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-dataset_info:
-  features:
-  - name: session_id
-    dtype: string
-  - name: logs
-    list: string
-  - name: metadata
-    dtype: string
-  - name: label
-    dtype: int64
-  - name: attack_ratio
-    dtype: float64
-  - name: severity_labels
-    dtype: string
-  - name: data_type
-    dtype: string
-  - name: is_attack
-    dtype: bool
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 1804071139
-    num_examples: 1704506
-  download_size: 250820904
-  dataset_size: 1804071139
----
+# LogAtlas-Defense-Set 🛡️🦊
 
-# ait-distilation-set
+A heterogeneous, labeled log dataset designed for training and evaluating log-level and session-level classifiers that distinguish between normal behavior and cyberattacks across multiple sources (system, network, and application logs). It is intended as the “defense layer” of the LogAtlas ecosystem, focusing on robust, realistic attack detection under varied class distributions.
 
-Labeled log sessions dataset with attack information.
+## Mascot
 
-## Dataset Description
+![ChatGPT Image Dec 14, 2025, 08_24_58 PM](https://cdn-uploads.huggingface.co/production/uploads/63fe5a071193ba2c229c29bf/vkh9n-MMEW2q2IGHGIznp.png)
 
-This dataset combines log sessions from `chYassine/log-sessions-supervised` with attack labels from `chYassine/ait-attack`.
+The LogAtlas-Defense-Set mascot is a vigilant cyber guardian fox in lightweight futuristic armor, standing on a shield made of interlocking log fragments with threat heatmaps. Its glowing tail traces highlight red “attack” regions in the log landscape, symbolizing precise and efficient detection of malicious activity in heterogeneous logs.
 
-### Dataset Structure
+## Dataset Overview
 
-Each session contains:
-- **session_id**: Unique identifier for the session
-- **logs**: List of log entries (JSON strings)
-- **metadata**: Session metadata including log types, duration, host info
-- **log_types**: Types of logs in the session
-- **has_attack**: Boolean indicating if session contains attacks
-- **attack_types**: List of attack types found in the session
-- **attack_count**: Number of logs with attacks
-- **attack_ratio**: Ratio of attack logs to total logs
-
-### Statistics
-
-- Total sessions: 1703363
-- Sessions with attacks: 1676374 (98.4%)
-- Sessions without attacks: 26989
-
-### Log Types
-
-- intranet.flores.wheeler.biz-access.log: 345282
-- intranet.hallbrown.wilson.com-access.log: 342532
-- intranet.mannsmith.harrison.com-access.log: 332309
-- intranet.price.fox.org-access.log: 328655
-- dnsmasq: 301906
-- apache2: 12666
-- audit: 7838
-- intranet.smith.santos.com-access: 6215
-- intranet.smith.russellmitchell.com-access.log: 6136
-- dnsteal: 5464
-
-### Attack Types
-
-- attacker_http,foothold,dirb: 1351557
-- dnsteal,attacker,dnsteal-received: 298307
-- attacker_http,foothold,wpscan: 22120
-- dns_scan,foothold: 2571
-- service_scan,foothold: 546
-- network_scan,foothold: 358
-- escalate,crack_passwords: 264
-- attacker_vpn,foothold: 181
-- attacker_http,foothold,webshell_cmd: 104
-- webshell_cmd,escalate: 75
-
-## Usage
-
-```python
-from datasets import load_dataset
-
-# Load the dataset
-dataset = load_dataset("chYassine/ait-distilation-set", split="train")
-
-# Example: Filter sessions with attacks
-attacks_only = dataset.filter(lambda x: x["has_attack"])
-
-# Example: Get sessions by log type
-dns_sessions = dataset.filter(lambda x: "dns" in x.get("log_types", []))
-```
-
-## Source Datasets
-
-- Sessions: [chYassine/log-sessions-supervised](https://huggingface.co/datasets/chYassine/log-sessions-supervised)
-- Attacks: [chYassine/ait-attack](https://huggingface.co/datasets/chYassine/ait-attack)
-
-## Citation
-
-If you use this dataset, please cite the original datasets as well.
+- **Goal**: Supervised training and evaluation of models that classify log sessions as normal or attack and support downstream severity-aware analysis.
+- **Scope**: Multi-source logs (e.g., system, network, application) organized into sessions with rich metadata, including attack ratios and severity labels.
+- **Use cases**:
+  - Training LLM-based or transformer-based detectors
+  - Benchmarking attack detection under different class distributions
+  - Studying severity-aware or source-aware threat modeling