TUKE-KEMT/reranking-skquad

This dataset is designed for training and evaluating reranking models for Slovak information retrieval.

It is structured to provide, for each query, a set of relevant (positive) and irrelevant (negative) documents. This format is particularly useful for evaluation of cross-encoder models.

The dataset is generated from three source files:

• A queries.jsonl file containing user queries.
• A corpus.jsonl file containing a large set of documents.
• A qrels.jsonl file that provides human-judged relevance scores linking queries to documents in the corpus.

Each example in the final dataset consists of a query, a list of positive passages, and a list of negative passages. A passage is considered "positive" if its relevance score is 4 or higher, and "negative" otherwise.

Example Data Instance:

{
  "query": "Aká je hlavná funkcia mitochondrií v bunke?",
  "positives": [
    "Mitochondrie sú známe ako 'elektrárne' bunky. Ich hlavnou funkciou je generovanie väčšiny bunkovej zásoby adenozíntrifosfátu (ATP), ktorý sa používa ako zdroj chemickej energie."
  ],
  "negatives": [
    "Jadro bunky obsahuje genetický materiál.",
    "Ribozómy sú zodpovedné za syntézu proteínov.",
    "Bunka je základnou stavebnou a funkčnou jednotkou všetkých známych organizmov."
  ]
}

Data Fields:

• query: A string representing the search query.
• positives: A list of strings, where each string is a document considered highly relevant to the query.
• negatives: A list of strings, where each string is a document considered less relevant to the query.

This dataset is ideal for tasks such as document reranking, answer passage selection, and semantic relevance classification.