Bhawna Piryani · Abdelrahman Abdallah · Jamshid Mozafari · Avishek Anand · Adam Jatowt
University of Innsbruck · TU Delft
📄 Read the Paper on arXiv | 🗓️ 2025
- 📘 Overview
- 📊 Datasets
- 🔧 Methods & Approaches
- 📖 Temporal Tasks
- 🏥 Domain-Specific Applications
- 🛠️ Resources & Tools
- 🚀 Future Directions
- 📝 Citation
This repository provides a comprehensive, curated collection of research papers, datasets, methods, and resources focused on Temporal Question Answering (TQA) and Temporal Information Retrieval (Temporal IR). It accompanies our survey paper on how AI models reason about time, adapt to evolving knowledge, answer temporally constrained questions, and retrieve time-sensitive information.
✨ Comprehensive Survey: Coverage of 27+ datasets, 50+ methods spanning 2003-2025
📊 Unified Taxonomy: Systematic categorization of tasks, datasets, and approaches
🔍 Critical Analysis: Evaluation of current capabilities and fundamental limitations
🚀 Research Roadmap: 7 critical directions for advancing temporal reasoning in AI
Time shapes how we:
- 🗞️ Retrieve information: "Latest climate policies" vs. "policies from the 1990s"
- 🧠 Reason about events: Understanding causality, change, and evolution
- 💬 Interact with AI: Expecting contextually appropriate temporal grounding
- 🔄 Adapt to change: Handling evolving facts and knowledge updates
- 27+ TQA Datasets covering diverse domains and temporal scopes
- 2.5M+ Questions spanning historical archives (1367) to real-time web (2025)
- Dataset Categories: Diachronic, Synchronic, Web-based, Synthetic, KG-based
🗞️ Diachronic Datasets (Time-Stamped Historical Documents)
| Dataset | Year | #Questions | Source | Time Coverage | Answer Type | Links |
|---|---|---|---|---|---|---|
| ArchivalQA | 2022 | 532K | NYT Corpus | 1987-2007 | Extractive | Paper · GitHub |
| ChroniclingAmericaQA | 2024 | 485K | Historical Newspapers | 1800-1920 | Extractive | Paper · GitHub |
| StreamingQA | 2022 | 147K | News Articles | 2007-2020 | Extractive | Paper · GitHub |
| NewsQA | 2017 | 119K | CNN/Daily Mail | 2007-2015 | Freeform | Paper · GitHub |
| TempLAMA | 2022 | 50K | News | 2010-2020 | Extractive | Paper · GitHub |
| TORQUE | 2020 | 21K | News | - | Abstractive | Paper · GitHub |
| ForecastQA | 2021 | 10.3K | News | 2015-2019 | Multiple Choice | Paper · Website |
| TDDiscourse | 2019 | 6.1K | News | Unspecified | Extractive | Paper · GitHub |
📖 Synchronic Datasets (Wikipedia Snapshots)
| Dataset | Year | #Questions | Time Scope | Answer Type | Multi-Hop | Links |
|---|---|---|---|---|---|---|
| ComplexTempQA | 2024 | 100.2K | 1987-2023 | Extractive | ✓ | Paper · GitHub |
| TEMPREASON | 2023 | 52.8K | 634-2023 | Abstractive | ✗ | Paper · GitHub |
| TimeQA | 2021 | 41.2K | 1367-2018 | Extractive | ✗ | Paper · GitHub |
| TemporalAlignmentQA | 2024 | 20K | 2000-2023 | Abstractive | ✗ | Paper Github |
| SituatedQA | 2021 | 12.2K | ≤ 2021 | Mixed | ✗ | Paper · GitHub |
| TempTabQA | 2023 | 11.4K | Infoboxes | Abstractive | ✗ | Paper · Website |
| TiQ | 2024 | 10K | Unspecified | Entities | ✗ | Paper · GitHub |
| PAT-Questions | 2024 | 6.1K | Present-anchored | Extractive | ✓ | Paper · GitHub |
| TRACIE | 2021 | 5.4K | ≤ 2020 | Abstractive | ✗ | Paper · GitHub |
| MenatQA | 2023 | 2.8K | 1367-2018 | Extractive | ✗ | Paper · GitHub |
🌐 Web & Real-Time Datasets
| Dataset | Year | #Questions | Source | Update Frequency | Links |
|---|---|---|---|---|---|
| ReaLTimeQA | 2023 | 5.1K | Web Search | Weekly (2020-2024) | Paper · Website |
| FreshQA | 2024 | 600 | Google Search | Periodic | Paper · GitHub |
🧪 Synthetic & Reasoning-Focused Datasets
| Dataset | Year | #Questions | Focus | Links |
|---|---|---|---|---|
| COTEMPQA | 2024 | 4.7K | Co-temporal reasoning | Paper · GitHub |
| UnSeenTimeQA | 2024 | 3.6K | Beyond memorization | Paper · GitHub |
| Test of Time (ToT) | 2024 | 1.8K | Temporal reasoning eval | Paper · GitHub |
| TIMEDIAL | 2021 | 1.1K | Temporal commonsense | Paper · GitHub |
📅 2003-2010: Rule-Based Era
└─ TimeML, TERSEO, temporal taggers
📅 2011-2019: Statistical & Early Neural
└─ Language models, temporal embeddings
📅 2020-2022: Transformer Revolution
└─ Temporal pretraining, time-aware architectures
📅 2023-2025: LLM & RAG Era
└─ Retrieval-augmented generation, temporal reasoning
🤖 Temporal Language Models (Click to expand all models)
| Model | Year | Key Innovation | Architecture | Paper | Code |
|---|---|---|---|---|---|
| TempoT5 | 2022 | Temporal conditioning via prefixes | T5 + timestamp prefixes | Paper | GitHub |
| BiTimeBERT | 2023 | Dual temporal encoding (timestamp + content) | BERT + bi-temporal module | Paper | Github |
| TempoBERT | 2022 | Time-aware masking strategy | BERT + temporal masking | Paper | GitHub |
| TALM | 2023 | Hierarchical temporal word representations | BERT + temporal adapter | Paper | Github |
| SG-TLM | 2023 | Syntax-guided + temporal-aware masking | BERT + dual masking | Paper | GitHub |
| TSM | 2023 | Temporal span masking | T5 + salient span masking | Paper | Contact authors |
| Temporal Attention | 2022 | Time matrix in attention mechanism | Transformer + time matrix | Paper | GitHub |
| TCQA | 2023 | Synthetic QA + span selection | T5-based | Paper | Github |
| Time-aware Prompting | 2022 | Temporal prompts for generation | GPT-2 + temporal prompts | Paper | GitHub |
🔍 Temporal RAG Systems (Click to expand all systems)
| System | Year | Pipeline Architecture | Temporal Signals | Paper | Code |
|---|---|---|---|---|---|
| TempRetriever | 2025 | Fusion-based dense retrieval | Query + doc timestamps | Paper | Contact authors |
| TimeR4 | 2024 | Retrieve-Rewrite-Retrieve-Rerank | TKG timestamps + constraints | Paper | GitHub |
| MRAG | 2024 | Modular multi-hop framework | Symbolic + semantic temporal scoring | Paper | Contact authors |
| TempRALM | 2024 | Dense retrieval + temporal proximity | Timestamp-based ranking | Paper | Contact authors |
| TsContriever | 2024 | Contrastive time-sensitive retrieval | Time-aware embeddings | Paper | Github |
| FreshLLMs | 2024 | Search augmentation for recency | Web search integration | Paper | GitHub |
🧠 Temporal Reasoning Methods (Click to expand all approaches)
| Method | Year | Reasoning Type | Key Contribution | Paper | Code |
|---|---|---|---|---|---|
| ECONET | 2021 | Continual adaptation | Event consistency across updates | Paper | GitHub |
| ConTempo | 2024 | Contrastive temporal relations | Unified temporal relation extraction | Paper | GitHub |
| TIMERS | 2021 | Document-level relations | Structured inference layers | Paper | GitHub |
| TRAM | 2024 | Multi-dimensional reasoning | Event frequency, duration, ordering | Paper | GitHub |
| TODAY | 2023 | Differential analysis | Temporal robustness testing | Paper | GitHub |
| Narrative-of-Thought | 2024 | Narrative-based reasoning | Recounted narratives for coherence | Paper | GitHub |
📜 Classical Methods (Rule-Based & Statistical)
| Era | Methods | Key Papers |
|---|---|---|
| Rule-Based | TimeML, TERSEO, temporal taggers | Harabagiu & Bejan, 2005, Saquete et al., 2004, Saquete et al., 2004 |
| Statistical IR | Time-based language models, temporal ranking | Li & Croft, 2003, Berberich et al., 2010, Arikan et al., 2009, Alonso et al., 2007, , , |
Core temporal prediction tasks supporting TQA systems:
| Task | Input | Output | Key Applications | Representative Papers |
|---|---|---|---|---|
| Event Dating | Event description | Event timestamp | Historical analysis, timeline construction | Das et al., 2017, Wang et al., 2021 |
| Document Dating | Document text | Creation date | Digital preservation, metadata recovery | Kumar et al., 2012, Niculae et al. 2014, Vashishth et al. 2018, Jatowt et al. 2007, SalahEldeen and Nelson, 2013 |
| Focus Time Estimation | Document content | Discussed time period | Historical QA, event-centric retrieval | Jatowt et al., 2013, Jatowt et al., 2013, Shrivastava et al., 2017 |
| Query Time Profiling | Search query | Temporal intent/distribution | Time-aware search, query understanding | Kanhabua & Nørvåg, 2010,Jones and Diaz 2007 Dakka et al., 2008, Gupta and Berberich 2014 |
Challenges: Patient timeline reconstruction, symptom progression, treatment sequencing
| System/Dataset | Focus | Key Paper |
|---|---|---|
| TimeText | Time-oriented clinical QA | Zhou et al., 2008 |
| Temporal Clinical QA | Semantic web techniques | Tao et al., 2010 |
| Time-aware Health QA | Evidence retrieval with recency | Vladika & Matthes, 2024 |
Challenges: Evolving statutes, precedent timelines, jurisdiction-specific temporal expressions
| System/Dataset | Focus | Key Paper |
|---|---|---|
| ChronosLex | Time-aware incremental training | T.y.s.s et al., 2024 |
Challenges: Regulatory changes, market events, time-sensitive numerical reasoning
| Dataset | Focus | Key Paper |
|---|---|---|
| FinQA | Numerical reasoning over financial data | Chen et al., 2021 |
| FinTextQA | Long-form financial QA | Chen et al., 2024 |
| FinDER | Financial QA with RAG | Choi et al., 2025 |
| Tool | Year | Languages | Type | Features | Link |
|---|---|---|---|---|---|
| HeidelTime | 2010 | 200+ | Rule-based | High precision, domain adaptation | Paper · GitHub |
| SUTime | 2012 | English | Rule-based | Stanford CoreNLP integration | Paper · Website |
| CogCompTime | 2018 | English | Neural | Compositional temporal understanding | Paper · GitHub |
| Temponym Tagger | 2016 | English | Hybrid | Implicit temporal references | Paper |
| Collection | Period | Size | Domain | Access |
|---|---|---|---|---|
| NYT Annotated Corpus | 1987-2007 | 1.8M articles | News | LDC License |
| Chronicling America | 1800-1920 | Historical | Newspapers | Free Access |
| Newswire Corpus | 1878-1977 | 2.7M articles | News | HuggingFace |
| Wikipedia Dumps | Various | TB-scale | Encyclopedia | Wikimedia |
- Temporal Robustness Testing: Wallat et al., 2024
- TimeBench: Comprehensive temporal reasoning benchmark (Chu et al., 2024)
- TRAM: Multi-dimensional temporal reasoning evaluation (Wang & Zhao, 2024)
Our survey identifies 7 critical research areas requiring immediate attention:
Problem: Static corpora can't handle evolving facts
Challenge: Temporal propagation when updating related events
Needed: Real-time knowledge graphs with dependency tracking
Problem: LLMs hallucinate temporal information
Challenge: Resolving "last Tuesday" or "since our last chat"
Needed: Timeline memory, temporal reference resolution
Problem: Most systems use only one knowledge type
Challenge: Aligning historical trends with current snapshots
Needed: Cross-source temporal alignment algorithms
Problem: Systems treat all dates as exact
Challenge: "Around 476 AD", "mid-20th century"
Needed: Probabilistic temporal representations
Problem: Most work is English text-only
Challenge: Lunar calendars, visual time cues, cultural references
Needed: Cross-lingual temporal taggers, vision-language models
Problem: Many questions hide their time constraints
Challenge: Inferring "now" vs. "historically" from context
Needed: Context-dependent temporal intent detection
Problem: Standard metrics don't capture temporal coherence
Challenge: Measuring temporal grounding, not just accuracy
Needed: Temporal-aware evaluation protocols
If you find this work useful, please cite 📜our paper:
Piryani, B., Abdullah, A., Mozafari, J., Anand, A., & Jatowt, A. (2025). It's High Time: A Survey of Temporal Question Answering. arXiv preprint arXiv:2505.20243.
@article{piryani2025s,
title={It's High Time: A Survey of Temporal Question Answering},
author={Piryani, Bhawna and Abdullah, Abdelrahman and Mozafari, Jamshid and Anand, Avishek and Jatowt, Adam},
journal={arXiv preprint arXiv:2505.20243},
year={2025}
}
This project is licensed under the MIT License - see the LICENSE file for details.
We welcome contributions to keep this survey comprehensive and up-to-date!
If we've missed your work or you know of a relevant paper/dataset that should be included, please send us an email at:
Please include:
- Paper title and authors
- Link to paper and code/data (if available)
- Brief description of the contribution
You can also open an issue on GitHub.