Update README.md

README.md

@@ -43,14 +43,16 @@ Use this model to classify transcript windows (typically ~20 caption snippets) a
 
 ## Training data
 
-Trained on weak-labeled YouTube podcast transcript segments:
+## Training data
+
+The model was trained on a mix of real and synthetic transcript windows:
 
-- Positive examples: segments containing sponsor keywords/brands
-- Negative examples: normal podcast content
-- Base set: ~800 labeled samples (400 positive, 400 negative)
-- Augmented with synthetic variants via paraphrasing
+- **Base set (~800 samples):** weak-labeled YouTube podcast segments
+  - Positive: segments matching sponsor keywords/brands
+  - Negative: normal podcast content
+- **Augmented set (~1,200 samples):** synthetic variants generated with Llama 8B via Ollama, preserving the original label. Synthetic data generated through strategies; paraphrase, new scenario, style shift, fragment, vocabulary shift
 
-Labels are heuristic/weak — the model learns from keyword-labeled examples and may not generalise perfectly to all sponsor formats.
+Original labels are heuristic — the model learns from keyword-labeled examples. Synthetic data increases linguistic diversity but inherits the same label assumptions.
 
 ## Usage