ProfessorBob/text-embedding-dataset

Text embedding Datasets

The text embedding datasets consist of several (query, passage) paired datasets aiming for text-embedding model finetuning. These datasets are ideal for developing and testing algorithms in the fields of natural language processing, information retrieval, and similar applications.

Dataset Details

Each dataset in this collection is structured to facilitate the training and evaluation of text-embedding models. The datasets are diverse, covering multiple domains and formats. They are particularly useful for tasks like semantic search, question-answering systems, and document retrieval.

[MOOC MCQ Queries]

The "MOOC MCQ Queries" dataset is derived from FUN MOOC, an online platform offering a wide range of French courses across various domains. This dataset is uniquely valuable for its high-quality content, manually curated to assist students in understanding course materials better.

Content Overview:

• Language: French
• Domains:
  • History: 57 examples
  • Religion: 125 examples
  • [Other domains to be added]
• Dataset Description: Each record in the dataset includes the following fields:

  {
    "query_id": "Unique identifier for each query",
    "query": "Text of the multiple-choice question (MCQ)",
    "answers": ["List of correct answer choices"],
    "distractions": ["List of incorrect choices"],
    "relevant_docs": ["List of relevant document IDs aiding the answer"]
  }
  

• statistics:

  Category	Num. of Queries	Query Avg. Words	Number of Docs	Short Docs (<375 words)	Long Docs (≥375 words)	Doc Avg. Words
  history	57	11.31	224	147	77	351.79
  religion	125	15.08	126	78	48	375.63
  recherche	52	12.71	69	20	49	535.00
  python	85	21.24	194	27	167	552.60

[Wikitext generated Queries]

To complete

[Documents]

This dataset is an extensive collection of document chunkings or entire document for short texts, designed to complement the MOOC MCQ Queries and other datasets in the collection.

• chunking strategies:

  • MOOC MCQ Queries: documents are chunked according to their natural divisions, like sections or subsections, ensuring that each chunk maintains contextual integrity.

• content format:

  {
  "doc_id": "Unique identifier for each document",
  "doc": "Text content of the document"
  }