orgrctera/beir_cqadupstack_gaming

CQADupStack — Gaming (BEIR)

Dataset description

CQADupStack is a benchmark for community question answering (cQA) research built from Stack Exchange forums. It was introduced by Hoogeveen, Verspoor, and Baldwin at ADCS 2015. The collection spans twelve topical subcommunities (tags); each subforum is distributed with duplicate-question annotations derived from how users and moderators linked related threads. That makes CQADupStack a standard resource for studying duplicate question retrieval and related tasks over noisy, real-world Q&A text.

BEIR (Benchmarking IR, Thakur et al., NeurIPS 2021) repackages CQADupStack—along with many other public corpora—as a unified zero-shot information retrieval benchmark. BEIR treats each CQADupStack subforum (including Gaming) as a separate retrieval setting: a query is a question post, and relevant documents are other question posts marked as duplicates (or in the duplicate closure), evaluated against a corpus of posts from the same domain.

This repository (orgrctera/beir_cqadupstack_gaming) holds the Gaming slice in Parquet form for retrieval pipelines. Rows correspond to evaluation queries with relevance judgments (expected_output) pointing at corpus document identifiers from the BEIR cqadupstack-gaming corpus. The released split is test (1,595 examples), consistent with BEIR’s standard test evaluation for this subset.

Domain

The Gaming forum covers video games, platforms (e.g., PC/console), mechanics, compatibility, and performance—typical Stack Exchange Arqade–style content. Language is English.

Task: retrieval (CQADupStack Gaming)

The task is duplicate-question retrieval formulated as ad hoc retrieval:

1. Input: a natural-language question (the query).
2. Output: a ranked list of document IDs from the Gaming corpus (or scores over the full collection), such that documents judged relevant in the official qrels receive high rank.

Evaluation uses standard IR metrics (e.g., nDCG@k, Recall@k, MRR), as in BEIR’s evaluation scripts or compatible toolkits (Pyserini, ir-measures, etc.).

Note: Full retrieval evaluation requires the corpus (passages keyed by ID). Align expected_output document IDs with the same BEIR cqadupstack-gaming corpus you index for search.

Data format (this repository)

Field	Description
id	UUID for this example row.
input	The query text (question).
expected_output	JSON string: list of objects {"id": "<corpus-doc-id>", "score": <relevance>}. Scores follow BEIR qrels (typically 1 for relevant in binary settings).
metadata.query_id	Original BEIR / CQADupStack Gaming query identifier (string).
metadata.split	Split name; this dataset provides test.

Example 1

{
  "id": "bd7ad6eb-d9af-4f9e-b48b-2458e9e61afd",
  "input": "Are the XBLA and PC versions of Minecraft compatible?",
  "expected_output": "[{\"id\": \"143450\", \"score\": 1}]",
  "metadata.query_id": "65828",
  "metadata.split": "test"
}

Example 2

{
  "id": "99cb0b75-167c-47d9-a103-381d8e4930ec",
  "input": "How to speed up minecraft on my computer",
  "expected_output": "[{\"id\": \"14876\", \"score\": 1}, {\"id\": \"24836\", \"score\": 1}]",
  "metadata.query_id": "42307",
  "metadata.split": "test"
}

References

Original CQADupStack resource

Doris Hoogeveen, Karin M. Verspoor, Timothy Baldwin
CQADupStack: A Benchmark Data Set for Community Question-Answering Research
Proceedings of the 20th Australasian Document Computing Symposium (ADCS 2015), pages 3:1–3:8.

• DOI: 10.1145/2838931.2838934
• Original data and documentation: University of Melbourne NLP resources — CQADupStack

The paper presents the full multi-forum benchmark, duplicate annotations, and evaluation protocols for community QA research.

BEIR benchmark (CQADupStack as a retrieval task)

Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, Iryna Gurevych
BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models
NeurIPS 2021 (Datasets and Benchmarks Track).

Abstract (from arXiv): “Existing neural information retrieval (IR) models have often been studied in homogeneous and narrow settings, which has considerably limited insights into their out-of-distribution (OOD) generalization capabilities. To address this, and to facilitate researchers to broadly evaluate the effectiveness of their models, we introduce Benchmarking-IR (BEIR), a robust and heterogeneous evaluation benchmark for information retrieval. We leverage a careful selection of 18 publicly available datasets from diverse text retrieval tasks and domains and evaluate 10 state-of-the-art retrieval systems including lexical, sparse, dense, late-interaction and re-ranking architectures on the BEIR benchmark. Our results show BM25 is a robust baseline and re-ranking and late-interaction-based models on average achieve the best zero-shot performances, however, at high computational costs. In contrast, dense and sparse-retrieval models are computationally more efficient but often underperform other approaches, highlighting the considerable room for improvement in their generalization capabilities.”

• Paper: arXiv:2104.08663 — OpenReview
• Code and dataset index: BEIR on GitHub

Related Hugging Face resources

• Raw BEIR-style layouts (corpus / queries / qrels) appear in community mirrors such as BeIR/cqadupstack-* datasets.
• ir-datasets packaging: ir-datasets BEIR documentation for programmatic access to BEIR collections.

Citation

If you use CQADupStack, cite Hoogeveen et al. (ADCS 2015). If you use the BEIR packaging or evaluation protocol, also cite Thakur et al. (NeurIPS 2021). BibTeX for BEIR is available from the official repository.

License

Stack Exchange content is typically shared under Creative Commons terms (e.g., CC BY-SA). This card uses cc-by-sa-4.0 as a common designation for Stack Exchange–derived text; confirm against your corpus snapshot and Stack Exchange’s current licensing terms if strict compliance is required.

---

Dataset card maintained for the orgrctera/beir_cqadupstack_gaming Hub repository.