---
title: "Labels without labellers for financial documents"
description: "Pseudo-labeling, label noise and weakly supervised learning appear on nearly 500 recent papers, almost all in AI journals, and only 7 of them concern financial, legal or business documents. Banks and auditors have millions of unlabelled contracts, invoices and filings and few labelled ones. Can weak supervision from rules and models replace hand labelling, and what errors does it introduce?"
url: https://scholar4ai.com/ideas/weak-supervision-for-financial-documents
markdown_url: https://scholar4ai.com/ideas/weak-supervision-for-financial-documents.md
license: CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)
share_text: "Labels without labellers for financial documents: Pseudo-labeling, label noise and weakly supervised learning appear on nearly 500 recent papers, almost all in AI journals, and only 7 of them concern financial, legal or business documents."
date: "2026-10-23"
updated: "2026-10-23"
status: "seed"
tags: ["ai", "finance", "documents"]
keywords: ["weakly supervised learning", "label noise", "pseudo-labeling"]
---

# Labels without labellers for financial documents

Pseudo-labeling, label noise and weakly supervised learning appear on nearly 500 recent papers, almost all in AI journals, and only 7 of them concern financial, legal or business documents. Banks and auditors have millions of unlabelled contracts, invoices and filings and few labelled ones. Can weak supervision from rules and models replace hand labelling, and what errors does it introduce?

## What prompted this

Since October 2025, 489 papers carry a weak-supervision, pseudo-label, noisy-label or label-noise keyword (487 in AI journals): "pseudo-labeling" on 215, "label noise" on 99, "weakly supervised learning" on 62, "pseudo-label refinement" on 36. Only 7 of the 489 also carry a finance, document, text-classification, legal, contract, invoice, economics, accounting or news keyword.

## Questions

- For tasks such as classifying contract clauses, flagging unusual invoices and tagging risk factors in filings, how close does weak supervision from rules and large models get to hand-labelled accuracy, at what cost?
- Are the errors of weakly supervised models concentrated in the rare cases that matter most (fraud, non-standard clauses)?
- Do language models used as labellers reproduce their own biases in the labels, and how can that be audited?

## Data that could help

Public financial document datasets with partial labels (contract clause datasets, SEC filings), and a partner firm's documents with a small gold-labelled set.

## Next step

Label a contract clause dataset with rules and a language model, train on the weak labels, and compare accuracy on a gold set with a model trained on the same number of hand labels.

## Related research activity

*The keywords this note follows, in the tracked finance, accounting, economics, management, operations research and AI journals as of 2026-10-26*

| Keyword | Papers, last 30 days | Rank | Papers per month, Oct 23 – Oct 26* | Papers, last 36 months |
|---|---:|---:|---|---:|
| weakly supervised learning | 10 | #121 | 6, 3, 6, 2, 14, 7, 1, 2, 8, 4, 3, 9, 7, 6, 5, 9, 8, 6, 5, 5, 6, 11, 8, 3, 3, 4, 7, 6, 3, 9, 7, 8, 2, 10, 7, 6, 9 | 216 |
| label noise | 4 | #525 | 6, 9, 10, 16, 7, 9, 9, 8, 6, 8, 8, 14, 4, 6, 7, 9, 8, 9, 6, 14, 11, 11, 11, 11, 7, 8, 15, 16, 7, 23, 9, 15, 8, 8, 8, 5, 4 | 346 |
| pseudo-labeling | 8 | #188 | 17, 15, 13, 17, 18, 15, 7, 9, 16, 22, 17, 12, 13, 14, 16, 30, 12, 18, 23, 14, 15, 26, 19, 18, 16, 13, 23, 29, 26, 35, 23, 25, 15, 11, 21, 18, 7 | 651 |

## Cite this page

Scholar4AI. (2026). Labels without labellers for financial documents. Journal Research Rankings. Retrieved October 23, 2026, from https://scholar4ai.com/ideas/weak-supervision-for-financial-documents

```bibtex
@misc{scholar4ai-ideas-weak-supervision-for-financial-documents,
  author = {{Scholar4AI}},
  title = {Labels without labellers for financial documents},
  year = {2026},
  howpublished = {\url{https://scholar4ai.com/ideas/weak-supervision-for-financial-documents}},
  urldate = {2026-10-23},
  note = {Journal Research Rankings; data from OpenAlex (CC0); licensed CC BY-NC 4.0}
}
```

License: CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/).