Datasets:
paper_id stringlengths 9 58 | openreview_id stringlengths 9 58 | title stringlengths 5 214 ⌀ | conference stringclasses 4
values | year int64 2.01k 2.03k | pdf_sha256 stringlengths 64 64 | text stringlengths 1 1.3M | text_sha256 stringlengths 64 64 | source_text_sha256 stringlengths 64 64 | cleanup_version stringclasses 1
value | flag_count int64 0 2.73k | stats_json stringlengths 73 506 |
|---|---|---|---|---|---|---|---|---|---|---|---|
0UK8c2kg7c | 0UK8c2kg7c | InstructAV: Instruction Fine-tuning Large Language Models for Authorship Verification | colm | 2,024 | f71db4adab0775a79b42bb60f90d3ec2ef26f36d7a9806806b4db63b7b12623f | InstructAV: Instruction Fine-tuning Large Language Models
for Authorship Verification
Yujia Hu , ∗ Zhiqiang Hu , ∗ Chun-Wei Seah, Roy Ka-Wei Lee Information Systems Technology and Design Singapore University of Technology and Design
{yujia hu, chunwei seah, roy lee}@sutd.edu.sg
{zhiqiang hu}@mymail.sutd.edu.sg
Abstract... | 79526e561763915022cf7f06ecbe9ab9c70264f2841a322580e89f4a372711f9 | f44b2e21cbdf0b83ace86fb37ed7bf87eb65c49a5eda54fa15c6a912b3fcf014 | positioned-clean-v2 | 10 | {"removed_words": 144, "retained_words": 8614, "rules": {"dehyphenated": 15, "joined_hyphenated_wrap": 1, "page_number": 16, "repeated_review_header": 128}, "source_words": 8758} |
0VLBwQGWpA | 0VLBwQGWpA | ReAct Meets ActRe: Autonomous Annotation of Agent Trajectories for Contrastive Self-Training | colm | 2,024 | 3547184a04aed247ab08fa59e5b52a7f24962dfb5be57c4d0f2f2b9b56214df4 | ReAct Meets ActRe: Autonomous Annotation of Agent Trajectories for Contrastive Self-Training
Zonghan Yang 1 , Peng Li 2 ∗ , Ming Yan 3 , Ji Zhang 3 , Fei Huang 3 , Yang Liu 1,2,4 ∗
1 Department of Computer Science and Technology, Tsinghua University, Beijing, China 2 Institute for AI Industry Research (AIR), Tsinghua U... | a6bc8b85e14ce0ac2aad52b40091b6d4b7d8ac83d3e3fea4c07bd4544207e9e4 | da767f9135abb7ff128c4f8a6ecf3a705b4aac65a52d0e83750b3f68cb0f977f | positioned-clean-v2 | 15 | {"removed_words": 198, "retained_words": 10967, "rules": {"dehyphenated": 10, "joined_hyphenated_wrap": 5, "page_number": 22, "repeated_review_header": 176}, "source_words": 11165} |
0o95CVdNuz | 0o95CVdNuz | Effective Prompt Extraction from Language Models | colm | 2,024 | eb4b7c46e113145b14975f26c2ae77a01fa9940e922c74baead2f98b5c28f7dc | "Effective Prompt Extraction from Language Models\nYiming Zhang 1 ∗ Nicholas Carlini 2 Daphne Ippo(...TRUNCATED) | 4b18589e1cb366bd86a9f453708d7f254899f383252879199bfa8584acac02ae | 7a1b2193cf7d8bf0f055f55e86a7a2d6b77280641a264890d44b45a5ae104d13 | positioned-clean-v2 | 9 | "{\"removed_words\": 248, \"retained_words\": 12718, \"rules\": {\"combined_negation\": 6, \"compose(...TRUNCATED) |
0oiG1KigYN | 0oiG1KigYN | SPEER: Sentence-Level Planning of Long Clinical Summaries via Embedded Entity Retrieval | colm | 2,024 | f62737322b783a4106e5140ab77243486d9a5165e727cec2fc248a9bc03ada52 | "SPEER: Sentence-Level Planning of Long Clinical Summaries via Embedded Entity Retrieval\nGriffin Ad(...TRUNCATED) | b5b8f609b28befdd429929816d23ab338f354bfe500ad79f5db2d51b79bd9673 | 8ffd5e22bae8574119adbc6358f394ac4ad06249af8945b2eae95c85be2fd177 | positioned-clean-v2 | 13 | "{\"removed_words\": 180, \"retained_words\": 9693, \"rules\": {\"dehyphenated\": 11, \"joined_hyphe(...TRUNCATED) |
18iNTRPx8c | 18iNTRPx8c | See What LLMs Cannot Answer: A Self-Challenge Framework for Uncovering LLM Weaknesses | colm | 2,024 | f2b402243eb15627eb9d32ea3643c3fbb2790c5066c897918cb7078e22e3826f | "See What LLMs Cannot Answer: A Self-Challenge Framework for Uncovering LLM Weaknesses\nYulong Chen (...TRUNCATED) | 7634c7cbac38aeece27711f5a6cd38c75b58747f3fa491c83e0175aeee27229d | e81276b1cd7ae8a090e0a424a5dcaa83baf776154b4ef5b2da58370590d4bb14 | positioned-clean-v2 | 28 | "{\"removed_words\": 279, \"retained_words\": 15078, \"rules\": {\"dehyphenated\": 31, \"joined_hyph(...TRUNCATED) |
1Tny4KgGO2 | 1Tny4KgGO2 | From Strategic Narratives to Code-Like Cognitive Models: An LLM-Based Approach in A Sorting Task | colm | 2,024 | 0abb9f6ba3c6f0c326bcfdb18cebe625f31fd66875d4f929d601fa2ebf6d70a1 | "From Strategic Narratives to Code-Like Cognitive Models: An LLM-Based Approach in A Sorting Task\nH(...TRUNCATED) | 33d2a0c41c10a089dc5b9895936c3d3946ea324dbcac3aa8c6e8467bee01d139 | f7ecde0b32d8988e3a26998c89b5de5e2a84d6a99ba064a0e0660a89b2f07781 | positioned-clean-v2 | 8 | "{\"removed_words\": 127, \"retained_words\": 5271, \"rules\": {\"combined_negation\": 1, \"composed(...TRUNCATED) |
1ba209BACA | 1ba209BACA | Agent-DocEdit: Language-Instructed LLM Agent for Content-Rich Document Editing | colm | 2,024 | fb22ce7021bf6a6abcd3f88f51d8b96ffc0de7798a435807d030da7149d1d701 | "Agent-DocEdit: Language-Instructed LLM Agent for Content-\nRich Document Editing\nTe-Lin Wu 1 , ∗(...TRUNCATED) | 12ac987b7a9fe3feebc00260841971c16cef64adebde6a4664cc523ed7f1a6ae | 142819be9992f13b8e551c3678677913f4c5c6fde2315168228c8fd58c36398b | positioned-clean-v2 | 9 | "{\"removed_words\": 180, \"retained_words\": 9596, \"rules\": {\"dehyphenated\": 12, \"joined_hyphe(...TRUNCATED) |
1eg6UnpYu7 | 1eg6UnpYu7 | Prompt Public Large Language Models to Synthesize Data for Private On-device Applications | colm | 2,024 | 5a08a455614fd6906bc9b6eb8d1a6ac24a633c8fbdb6b19a735fbced1c462d61 | "Prompt Public Large Language Models to Synthesize Data for Private On-device Applications\nShanshan(...TRUNCATED) | df72d6fb4fadb876595b0ceeac5f9496801aa36c37517a9cd9222e1d1ca51cef | 3bfc810926aeaec1bef9321ea3dd6416e4d4084d6388cd2d9c7e9a1daa4961a9 | positioned-clean-v2 | 7 | "{\"removed_words\": 173, \"retained_words\": 10118, \"rules\": {\"dehyphenated\": 6, \"expanded_lig(...TRUNCATED) |
1pgfvZj0Rx | 1pgfvZj0Rx | "Risks from Language Models for Automated Mental Healthcare: Ethics and Structure for Implementation(...TRUNCATED) | colm | 2,024 | f9ba5a1701f0031a8ce80fd2715aaadbbf527c2f12971701adc968c04f3f8ff6 | "Risks from Language Models for Automated Mental Healthcare: Ethics and Structure for Implementation(...TRUNCATED) | d1be49a891c4f228e919b1bface13aa134320e0ac132f6520e7beff739517341 | d88da0c4c4b2719649cc32887bba984f8a99069aa6eacac0447f4a0ef72bd216 | positioned-clean-v2 | 22 | "{\"removed_words\": 351, \"retained_words\": 19183, \"rules\": {\"dehyphenated\": 21, \"joined_hyph(...TRUNCATED) |
2cop2jmQVL | 2cop2jmQVL | Stream of Search (SoS): Learning to Search in Language | colm | 2,024 | 2d8931191a7ce2bd15e1637fa64900d642898283324c07b383a034d35d1563de | "Stream of Search (SoS): Learning to Search in Language\nKanishk Gandhi ∗\nStanford University\nDe(...TRUNCATED) | d50a0a52db3fa7088ccc01fe651cfcf0250c4f569ee774ae5deca01e08885042 | 81251e890b19ecf6cf007e2574586899012986061b3774da75fe9fd62da0979a | positioned-clean-v2 | 5 | "{\"removed_words\": 154, \"retained_words\": 8865, \"rules\": {\"dehyphenated\": 6, \"joined_hyphen(...TRUNCATED) |
End of preview. Expand in Data Studio
Open Text Detector paper text
Full text of 145,377 machine-learning papers extracted from their PDFs: conference and journal submissions from OpenReview (accepted, rejected and withdrawn where public), used by Open Text Detector. Code: https://github.com/woog97/open-text-detector
Configs
clean(default): the text layer after cleanup (positioned-clean-v2: hyphenation, headers/footers, page numbers and similar layout artefacts removed).direct: the raw positioned text layer from poppler (pdftotext -bbox-layout), with page-level tesseract OCR only for pages without a usable text layer.methodsays which.omissions: the clean text with non-prose content replaced by markers: figure text, tables, display equations, garbled inline math, number runs and the reference list.omissions_jsonlists every omission with its type.
All three configs have one row per PDF, keyed by pdf_sha256 (SHA-256 of the PDF), with paper_id (the OpenReview
forum id), title, conference and year. Rows join across configs on pdf_sha256.
Contents
| Conference | Year | Extracted papers | Total papers |
|---|---|---|---|
| ICLR | 2027 | 42,344 (99.8%) | 42,429 |
| ICLR | 2026 | 19,732 (99.6%) | 19,814 |
| ICLR | 2025 | 11,618 (99.5%) | 11,672 |
| ICLR | 2024 | 7,304 (98.6%) | 7,404 |
| ICLR | 2023 | 4,824 (97.4%) | 4,955 |
| ICLR | 2022 | 3,363 (98.3%) | 3,422 |
| ICLR | 2021 | 2,973 (98.6%) | 3,014 |
| ICLR | 2020 | 2,559 (98.7%) | 2,594 |
| ICLR | 2019 | 1,553 (98.4%) | 1,579 |
| ICLR | 2018 | 987 (97.0%) | 1,018 |
| ICLR | 2017 | 486 (99.2%) | 490 |
| ICLR | 2014 | 69 (100.0%) | 69 |
| ICLR | 2013 | 63 (94.0%) | 67 |
| COLM | 2026 | 856 (100.0%) | 856 |
| COLM | 2025 | 418 (100.0%) | 418 |
| COLM | 2024 | 299 (100.0%) | 299 |
| CoRL | 2025 | 263 (100.0%) | 263 |
| CoRL | 2024 | 264 (100.0%) | 264 |
| CoRL | 2023 | 198 (99.5%) | 199 |
| CoRL | 2022 | 197 (100.0%) | 197 |
| CoRL | 2021 | 153 (100.0%) | 153 |
| EMNLP | 2023 | 2,018 (99.9%) | 2,020 |
| ICML | 2026 | 6,555 (100.0%) | 6,556 |
| ICML | 2025 | 3,422 (100.0%) | 3,423 |
| ICML | 2024 | 2,610 (100.0%) | 2,610 |
| ICML | 2023 | 1,828 (100.0%) | 1,828 |
| NeurIPS | 2025 | 6,130 (100.0%) | 6,131 |
| NeurIPS | 2024 | 4,761 (100.0%) | 4,762 |
| NeurIPS | 2023 | 3,730 (99.9%) | 3,733 |
| NeurIPS | 2022 | 2,958 (99.0%) | 2,987 |
| NeurIPS | 2021 | 2,689 (89.0%) | 3,020 |
| TMLR | 2026 | 3,027 (115.4%) | 2,623 |
| TMLR | 2025 | 2,228 (95.1%) | 2,344 |
| TMLR | 2024 | 1,498 (95.4%) | 1,571 |
| TMLR | 2023 | 924 (85.6%) | 1,079 |
| TMLR | 2022 | 476 (74.1%) | 642 |
Total papers: public submissions on OpenReview (accepted, rejected, withdrawn and desk rejected), from our catalogue of September 24, 2026 (ICLR 2027: October 4, 2026). The percentage is extracted papers over total papers.
ICLR 2027 papers are submissions under review at the time of extraction (October 2026).
Notes
- Extraction quality varies with the PDF; scanned or unusual PDFs can produce garbled text. Known open issues in the omissions config: small-caps headings next to tables, pseudocode lines omitted as math, untruled tables leaking cells.
- Each paper remains under the licence its authors chose; see the paper's OpenReview page. To ask for a paper to be removed, open a discussion on this dataset.
- Downloads last month
- 389