Upload README.md with huggingface_hub

README.md

@@ -1,30 +1,96 @@
 ---
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: test
-    path: data/test-*
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  - name: label
-    dtype:
-      class_label:
-        names:
-          '0': ham
-          '1': scam
-  - name: language
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 3256251
-    num_examples: 12600
-  - name: test
-    num_bytes: 361805
-    num_examples: 1400
-  download_size: 1642794
-  dataset_size: 3618056
+language:
+- as
+- bn
+- en
+- gu
+- hi
+- kn
+- ks
+- ml
+- mr
+- or
+- pa
+- ta
+- te
+tags:
+- scam-detection
+- fraud-detection
+- text-classification
+- indic
+- nlp
+- synthetic
+task_categories:
+- text-classification
+size_categories:
+- 10K<n<100K
+license: mit
+pretty_name: "Indian Multilingual Scam & Ham SMS Dataset (14 Languages)"
 ---
+
+# Indian Multilingual Scam & Ham SMS Dataset (14 Languages)
+
+A balanced benchmark dataset of **14,000 text samples** curated for detecting fraud, phishing, and legitimate messages across **14 major Indian languages and scripts**.
+
+## Dataset Summary
+- **Total Records:** 14,000 samples
+- **Label Split:** Exactly 7,000 Scam / 7,000 Ham (Balanced 50:50 distribution)
+- **Languages Covered (1,000 samples each across 14 languages):**
+  - Assamese (`as`)
+  - Bengali (`bn`)
+  - English (`en`)
+  - Gujarati (`gu`)
+  - Hindi (`hi`)
+  - Hinglish (`hi-Latn`)
+  - Kannada (`kn`)
+  - Kashmiri (`ks` - Perso-Arabic script)
+  - Malayalam (`ml`)
+  - Marathi (`mr`)
+  - Odia (`or`)
+  - Punjabi (`pa` - Gurmukhi script)
+  - Tamil (`ta`)
+  - Telugu (`te`)
+
+---
+
+## Dataset Structure
+
+### Features
+| Column | Type | Description |
+| :--- | :--- | :--- |
+| `text` | string | The full SMS / messaging alert text. |
+| `label` | ClassLabel | Target category: `0` (ham) or `1` (scam). |
+| `language` | string | Language name (e.g., `Hindi`, `Bengali`, `Odia`, `Hinglish`). |
+
+### Splitting
+- **Train:** 12,600 samples (90%)
+- **Test:** 1,400 samples (10%)
+- Stratified across both classes to preserve exact class balance.
+
+---
+
+## Domain Coverage & Fraud Vectors
+The dataset models realistic communication patterns across India:
+
+### 1. Scam Vectors
+- **Utility Threats:** Immediate power/electricity cut-off notices demanding urgent phone calls or payment.
+- **Banking / KYC Phishing:** Impersonation of major Indian public & private banks (SBI, PNB, HDFC) with fake KYC verification deadlines.
+- **Lottery / Scheme Baits:** Government subsidy lures (PM Awas Yojana), prize claims, and work visa processing fee scams.
+- **Delivery Frauds:** Fake package delivery delays requesting minor redelivery payments or credential confirmations.
+
+### 2. Ham (Legitimate) Samples
+- **Transactional Alerts:** Genuine OTP codes, bank debit/credit balance alerts, and telecom recharge receipts.
+- **Official Utility Reminders:** Real-world telecom and broadband billing dates.
+- **Conversational Messages:** Everyday peer-to-peer discussions, family updates, and social interactions in native scripts.
+
+---
+
+## Quick Usage
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("__anmolshrivastav/scam_ham_india_14_languages__")
+
+# Inspect a sample
+print(dataset["train"][0])