Marqo GS 10M
Multimodal dataset including text and product images, used to learn how to rank in search engines and e-commerce.
Approximately 9.8 million multimodal items (text + images), 212 GB, Parquet format
Apache 2.0
Description
The dataset Marqo GS 10M is a multimodal corpus of nearly 10 million examples, combining images and text descriptions of Google Shopping products. It is designed to train information search and ranking models using an innovative Generalized Contrastive Learning (GCL) method.
What is this dataset for?
- Improving the performance of multimodal product search engines
- Training ranking and information retrieval models in e-commerce
- Experiment with advanced multimodal contrasted learning techniques
Can it be enriched or improved?
Yes, by further annotating product categories, by adding metadata such as prices or customer reviews, or by creating specialized subsets according to market segments.
🔎 In summary
🧠 Recommended for
- Information researchers
- E-commerce developers
- Multimodal data scientists
🔧 Compatible tools
- PyTorch
- TensorFlow
- LangChain
- Apache Spark
- Pandas
💡 Tip
Exploit data in batches and combine with pre-trained embeddings to optimize learning.
Frequently Asked Questions
Does this dataset contain associated images and text?
Yes, each example has a text description and a product image.
Can this dataset be used to search for e-commerce information?
Yes, it's specifically designed to improve product ranking models.
What is the size of the dataset and in what format is it available?
The dataset is about 212 GB, stored in Parquet format, suitable for batch processing.
