Karimansour/Synthetic_IR

Synthetic IR Dataset

Description

This synthetic dataset is designed for training and evaluating Information Retrieval (IR) models, especially for document relevance classification tasks. It contains query-document pairs with binary relevance labels.

Queries and documents are generated for three distinct topics:

• sports: football, basketball, tennis, match, score
• tech: AI, machine learning, cloud, software, algorithm
• health: diet, exercise, nutrition, wellness, fitness

For each query, the dataset contains:

• A relevant document about the same topic as the query (label 1).
• A non-relevant document about a different topic (label 0).

This dataset can be used to train models that classify whether a document is relevant to a query or not.

Dataset Statistics

• Number of generated queries: 1000
• Number of query-document pairs: 2000 (each query paired with one relevant and one non-relevant document)
• Balanced binary labels: 50% relevant (1), 50% non-relevant (0)

Example Data Points

{
  "query": "What is football?",
  "document": "Sports is about basketball, tennis, match. It involves tennis.",
  "label": 1
}