Elamur Research⌕
☰
home / papers / 2026 / Fin-STAR: Structure-as-Semantics to Resolve Implicitness in Financial Retrieval

Fin-STAR: Structure-as-Semantics to Resolve Implicitness in Financial Retrieval

Paper · 2026 · Yu Zou, Yan Chen, Lida He et al.

Summary

Financial filings often leave key facts implicit. A figure or a generic term only becomes meaningful through section headers and document hierarchy that may sit far from the snippet itself. Standard RAG chunking loses this context. Summary-based hierarchical methods blur exact values, and graph-based RAG ignores which section a fact belongs to. Fin-STAR treats structure as part of a snippet's meaning rather than as a navigation aid.

During indexing, it builds self-contained snippets, keeping tables intact and linearizing them. A small language model then adds a "virtual node" summarizing each snippet to its header path. This path is injected through cross-attention, with the snippet acting as query. The result is concatenated with the original embedding so that entities and numbers stay unchanged. Online retrieval then plans sub-queries, gathers hybrid evidence and composes a verifiable answer.

On the expert-checked FinTierQA Gold benchmark, Fin-STAR beat GraphRAG, LightRAG and RAPTOR in most settings. The authors also release FinTierQA, a large tiered resource for structure-aware financial QA.

Key points

  • The evaluation covers 15 regimes (CN, HK and US markets × 5 difficulty tiers). Fin-STAR achieves the top Joint F1 in 12 of them, with baselines only marginally ahead in cases such as US-T1 and HK-T4.
  • FinTierQA contains 3,928,500 automatically generated QA pairs from over 78k filings (2015–2024). Each pair carries a question, answer, evidence and tier label, split from 1,178,550 (T1) down to 392,850 (T5).
  • FinTierQA Gold was verified by consensus among 18 recruited financial experts.
  • In the US ablation, removing virtual nodes drops T5 Joint F1 from 20.3 to 6.6. Replacing cross-attention with prefix concatenation, or dropping the factual-invariance constraint, also lowers scores.
  • Enriched structural paths are capped at depth 5. The authors recommend using the system as an analyst copilot, not for autonomous trading.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Southwestern University of Finance and EconomicsChengdu · China5 works · 1 citation

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 552 works

Cite

Cite as Yu Zou, Yan Chen, Lida He, Qi Zhou, Xiaorui Zhou, Aixi Zhong, Yi Wang, Wei Li, Qingyu Wang, Jiatao Li, Wei Gong, Jialei Zeng, Jingmei Zhao, Ke Jiang and Qing Li. 2026. Fin-STAR: Structure-as-Semantics to Resolve Implicitness in Financial Retrieval. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.2151

@inproceedings{zou2026fin,
  title = {{Fin-STAR: Structure-as-Semantics to Resolve Implicitness in Financial Retrieval}},
  author = {Yu Zou, Yan Chen, Lida He, Qi Zhou, Xiaorui Zhou, Aixi Zhong, Yi Wang, Wei Li, Qingyu Wang, Jiatao Li, Wei Gong, Jialei Zeng, Jingmei Zhao, Ke Jiang and Qing Li},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.2151},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.2151}
}

References

23 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Yan Chen, Yu Zou, Jialei Zeng, Haoran You, Xiaorui Zhou, and Aixi Zhong. 2025. Pharos-esg: A framework for multimodal parsing, contextual narration, and hierarchical labeling of esg report. arXiv preprint arXiv:2511.16417. — [arXiv:2511.16417](https://arxiv.org/abs/2511.16417) find
  2. Zhiyu Chen, Wenhu Chen, Charese Smiley, Sameena Shah, Iana Borova, Dylan Langdon, Reema Moussa, Matt Beane, Ting-Hao Huang, Bryan Routledge, and William Yang Wang. 2021. Finqa: A dataset of numerical reasoning over financial data. In Proceedings of the 2021 Conference on Empirical Methods in Natural…
  3. Wei-Lin Chiang, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Banghua Zhu, Hao Zhang, Michael Jordan, Joseph E Gonzalez, and 1 others. 2024. Chatbot arena: An open platform for evaluating llms by human preference. In Forty-first International Conference on Machin…
  4. Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Dasha Metropolitansky, Robert Osazuwa Ness, and Jonathan Larson. 2024. From local to global: A graph rag approach to query-focused summarization. arXiv preprint arXiv:2404.16130. find
  5. Joseph L Fleiss. 1971. Measuring nominal scale agreement among many raters. Psychological bulletin, 76(5):378. find
  6. Gauthier Guinet, Behrooz Omidvar-Tehrani, Anoop Deoras, and Laurent Callot. 2024. Automated evaluation of retrieval-augmented language models with task-specific exam generation. arXiv preprint arXiv:2405.13622. — [arXiv:2405.13622](https://arxiv.org/abs/2405.13622) find
  7. Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang. 2024. Lightrag: Simple and fast retrieval-augmented generation. arXiv preprint arXiv:2410.05779.
  8. Bernal Jiménez Gutiérrez, Yiheng Shu, Weijian Qi, Sizhe Zhou, and Yu Su. 2025. From rag to memory: Non-parametric continual learning for large language models. arXiv preprint arXiv:2502.14802.
  9. Pranab Islam, Anand Kannappan, Douwe Kiela, Rebecca Qian, Nino Scherrer, and Bertie Vidgen. 2023. Financebench: A new benchmark for financial question answering. arXiv preprint arXiv:2311.11944. — [arXiv:2311.11944](https://arxiv.org/abs/2311.11944) find
  10. Gautier Izacard and Edouard Grave. 2021. Leveraging passage retrieval with generative models for open domain question answering. In Proceedings of the 16th conference of the european chapter of the association for computational linguistics: main volume, pages 874–880.
  11. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick SH Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. In EMNLP (1), pages 6769–6781.
  12. Viet Lai, Michael Krumdick, Charles Lovering, Varshini Reddy, Craig Schmidt, and Chris Tanner. 2025. Sec-qa: A systematic evaluation corpus for financial qa. In Proceedings of The 10th Workshop on Financial Technology and Natural Language Processing, pages 221–236. find
  13. Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. 2016. Squad: 100,000+ questions for machine comprehension of text. arXiv preprint arXiv:1606.05250.
  14. Varshini Reddy, Rik Koncel-Kedziorski, Viet Dac Lai, Michael Krumdick, Charles Lovering, and Chris Tanner. 2024. Docfinqa: A long-context financial reasoning dataset. arXiv preprint arXiv:2401.06915. find
  15. Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D Manning. 2024. Raptor: Recursive abstractive processing for tree-organized retrieval. In The Twelfth International Conference on Learning Representations.
  16. Shu Wang, Yixiang Fang, Yingli Zhou, Xilin Liu, and Yuchi Ma. 2025. Archrag: Attributed community-based hierarchical retrieval-augmented generation. arXiv preprint arXiv:2502.09891. — [arXiv:2502.09891](https://arxiv.org/abs/2502.09891) find
  17. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D Manning. 2018. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 conference on empirical methods in natural language processing, pages 2369–…
  18. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. In The eleventh international conference on learning representations.
  19. Mingtian Zhang, Yu Tang, and PageIndex Team. 2025a. Pageindex: Next-generation vector-less, reasoning-based rag. PageIndex Blog. Https://pageindex.ai/blog/pageindex-intro. find
  20. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025b. Qwen3 embedding: Advancing text embedding and reranking through foundation models. arXiv preprint arXiv:2506.05176.
  21. Fengbin Zhu, Wenqiang Lei, Youcheng Huang, Chao Wang, Shuo Zhang, Jiancheng Lv, Fuli Feng, and Tat-Seng Chua. 2021. Tat-qa: A question answering benchmark on a hybrid of tabular and textual content in finance. arXiv preprint arXiv:2105.07624.
  22. Feng Li. 2008. Annual report readability, current earnings, and earnings persistence. Journal of Accounting and economics, 45(2-3):221–247. find
  23. Tim Loughran and Bill McDonald. 2011. When is a liability not a liability? textual analysis, dictionaries, and 10-ks. The Journal of finance, 66(1):35–65. *Figure 4: The standardized prompt used for the AI expert panel to generate annotations for the FinTierQA Gold benchmark.* find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zou-2026-fin-star-structure-as-semantics/