FutureQueryEval / README.md
abdoelsayed's picture
Update README.md
e57e5b2 verified
---
license: apache-2.0
task_categories:
- text-retrieval
language:
- en
tags:
- information-retrieval
- reranking
- temporal-evaluation
- benchmark
size_categories:
- 1K<n<10K
pretty_name: Reranking, Retreiver
---
# FutureQueryEval Dataset (EMNLP 2025)🔍
## Dataset Description
**FutureQueryEval** is a novel Information Retrieval (IR) benchmark designed to evaluate reranker performance on temporal novelty. It comprises **148 queries** with **2,938 query-document pairs** across **7 topical categories**, specifically created to test how well reranking models generalize to truly novel queries that were unseen during LLM pretraining.
### Key Features
- **Zero Contamination**: All queries refer to events after April 2025
- **Human Annotated**: Created by 4 expert annotators with quality control
- **Diverse Domains**: Technology, Sports, Politics, Science, Health, Business, Entertainment
- **Real Events**: Based on actual news and developments, not synthetic data
- **Temporal Novelty**: First benchmark designed to test reranker generalization on post-training events
## Dataset Statistics
| Metric | Value |
|--------|-------|
| Total Queries | 148 |
| Total Documents | 2,787 |
| Query-Document Pairs | 2,938 |
| Avg. Relevant Docs per Query | 6.54 |
| Languages | English |
| License | Apache-2.0 |
## Category Distribution
| Category | Queries | Percentage |
|----------|---------|------------|
| **Technology** | 37 | 25.0% |
| **Sports** | 31 | 20.9% |
| **Science & Environment** | 20 | 13.5% |
| **Business & Finance** | 19 | 12.8% |
| **Health & Medicine** | 16 | 10.8% |
| **World News & Politics** | 14 | 9.5% |
| **Entertainment & Culture** | 11 | 7.4% |
## Dataset Structure
The dataset consists of three main files:
### Files
- **`queries.tsv`**: Contains the query information
- Columns: `query_id`, `query_text`, `category`
- **`corpus.tsv`**: Contains the document collection
- Columns: `doc_id`, `title`, `text`, `url`
- **`qrels.txt`**: Contains relevance judgments
- Format: `query_id 0 doc_id relevance_score`
### Data Fields
#### Queries
- `query_id` (string): Unique identifier for each query
- `query_text` (string): The natural language query
- `category` (string): Topical category (Technology, Sports, etc.)
#### Corpus
- `doc_id` (string): Unique identifier for each document
- `title` (string): Document title
- `text` (string): Full document content
- `url` (string): Source URL of the document
#### Relevance Judgments (qrels)
- `query_id` (string): Query identifier
- `iteration` (int): Always 0 (standard TREC format)
- `doc_id` (string): Document identifier
- `relevance` (int): Relevance score (0-3, where 3 is highly relevant)
## Example Queries
**🌍 World News & Politics:**
> "What specific actions has Egypt taken to support injured Palestinians from Gaza, as highlighted during the visit of Presidents El-Sisi and Macron to Al-Arish General Hospital?"
**⚽ Sports:**
> "Which teams qualified for the 2025 UEFA European Championship playoffs in June 2025?"
**💻 Technology:**
> "What are the key features of Apple's new Vision Pro 2 announced at WWDC 2025?"
## Usage
### Loading the Dataset
```python
from datasets import load_dataset
# Load the dataset
dataset = load_dataset("abdoelsayed/FutureQueryEval")
# Access different splits
queries = dataset["queries"]
corpus = dataset["corpus"]
qrels = dataset["qrels"]
# Example: Get first query
print(f"Query: {queries[0]['query_text']}")
print(f"Category: {queries[0]['category']}")
```
### Evaluation Example
```python
import pandas as pd
# Load relevance judgments
qrels_df = pd.read_csv("qrels.txt", sep=" ",
names=["query_id", "iteration", "doc_id", "relevance"])
# Filter for a specific query
query_rels = qrels_df[qrels_df["query_id"] == "FQ001"]
print(f"Relevant documents for query FQ001: {len(query_rels)}")
```
## Methodology
### Data Collection Process
1. **Source Selection**: Major news outlets, official sites, sports organizations
2. **Temporal Filtering**: Events after April 2025 only
3. **Query Creation**: Manual generation by domain experts
4. **Novelty Validation**: Tested against GPT-4 knowledge cutoff
5. **Quality Control**: Multi-annotator review with senior oversight
### Annotation Guidelines
- **Highly Relevant (3)**: Document directly answers the query
- **Relevant (2)**: Document partially addresses the query
- **Marginally Relevant (1)**: Document mentions query topics but lacks detail
- **Not Relevant (0)**: Document does not address the query
## Research Applications
This dataset is designed for:
- **Reranker Evaluation**: Testing generalization to novel content
- **Temporal IR Research**: Understanding time-sensitive retrieval challenges
- **Domain Robustness**: Evaluating cross-domain performance
- **Contamination Studies**: Clean evaluation on post-training data
## Benchmark Results
Top performing methods on FutureQueryEval:
| Method | Type | NDCG@10 | Runtime (s) |
|--------|------|---------|-------------|
| Zephyr-7B | Listwise | **62.65** | 1,240 |
| MonoT5-3B | Pointwise | **60.75** | 486 |
| Flan-T5-XL | Setwise | **56.57** | 892 |
## Dataset Updates
FutureQueryEval will be updated every 6 months with new queries about recent events to maintain temporal novelty:
- **Version 1.1** (December 2025): +100 queries from July-September 2025
- **Version 1.2** (June 2026): +100 queries from October 2025-March 2026
## Citation
If you use FutureQueryEval in your research, please cite:
```bibtex
@misc{abdallah2025good,
title={How Good are LLM-based Rerankers? An Empirical Analysis of State-of-the-Art Reranking Models},
author={Abdelrahman Abdallah and Bhawna Piryani and Jamshid Mozafari and Mohammed Ali and Adam Jatowt},
year={2025},
eprint={2508.16757},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Contact
- **Authors**: Abdelrahman Abdallah, Bhawna Piryani
- **Institution**: University of Innsbruck
- **Paper**: [arXiv:2508.16757](https://arxiv.org/abs/2508.16757)
- **Code**: [GitHub Repository](https://github.com/DataScienceUIBK/llm-reranking-generalization-study)
## License
This dataset is released under the Apache-2.0 License.