MIRACL: Multilingual Information Retrieval Triplets for Embedding Training

A reformatted version of the MIRACL dataset used to train the BGE-M3 embedding model. The dataset was uploaded by sentence-transformers and last updated on June 20, 2024. It contains a subset structured as text triplets with anchor, positive, and negative examples.

Use Cases

• Training contrastive learning models based on the provided anchor-positive-negative triplet structure.
• Evaluating multilingual retrieval performance based on the dataset's cross-lingual information retrieval focus.
• Fine-tuning embedding models for semantic similarity tasks based on the positive and negative text pairs.

Strengths

• Structured for contrastive learning with explicit anchor, positive, and negative text columns.
• Designed for training a specific, known embedding model (BGE-M3), suggesting relevance for model development.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

huggingface

Collection Method

Reformatting of the original MIRACL dataset.

Time Range

null

Freshness

Last updated 2024-06-20 13:50:24; freshness should be verified.

Geography

null