| --- |
| license: mit |
| --- |
| # Vietnamese Embedding Training Data |
|
|
| A curated Vietnamese dataset for training semantic embedding and retrieval models. |
|
|
| This dataset combines multiple Vietnamese sources and normalizes them into a unified retrieval-oriented format with positive passages and hard negatives. |
|
|
| ## Format |
|
|
| Each sample contains: |
|
|
| ```json |
| { |
| "query": "...", |
| "positive": "...", |
| "negatives": ["...", "...", "..."], |
| "source": "...", |
| "task": "retrieval", |
| "domain": "...", |
| "group_id": "...", |
| "difficulty": "hard" |
| } |
| ``` |
|
|
| ## Processing |
|
|
| The dataset is processed with: |
|
|
| * Unicode and whitespace normalization |
| * Duplicate removal |
| * Basic quality filtering |
| * Easy-negative filtering |
| * False-negative filtering |
| * BM25-based hard-negative mining |
| * Source and domain metadata preservation |
|
|
| ## Intended Use |
|
|
| This dataset is designed for training: |
|
|
| * Vietnamese text embedding models |
| * Dense retrieval models |
| * Semantic search systems |
| * RAG retrieval models |
| * Matryoshka embedding models |
|
|
| It is especially suitable for contrastive learning with InfoNCE, in-batch negatives, and explicit hard negatives. |
|
|
