{"url":"/sota/long-form-narrative-summarization-on-booksum","task":{"name":"Long-Form Narrative Summarization","url":"/task/long-form-narrative-summarization","note":null},"dataset":{"name":"BookSum","url":"/dataset/booksum"},"category":"Natural Language Processing","categories":["Knowledge Base","Natural Language Processing"],"category_note":null,"description":"Summarizing long-form narratives, such as books,\r\nmovies, and TV scripts, remains an open challenge in NLP. Unlike news or document summarization, narratives require capturing intricate plotlines, evolving character relationships, and thematic coherence over tens of thousands of tokens. The hybrid structure of\r\nnarratives, which combines descriptive prose with\r\nmulti-speaker dialogues, implicit inference, and dynamic topic shifts, adds further complexity, demanding an approach that preserves contextual integrity\r\nwhile condensing information effectively. Furthermore, the sheer length of narrative texts, typically\r\nranging from 40K to 160K tokens, poses significant challenges for standard summarization models.","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["BERTScore (F1)","ROUGE-1","ROUGE-2","ROUGE-L","ROUGE (geometric mean of 1/2/L)"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"BERTScore (F1)":"higher","ROUGE-1":"higher","ROUGE-2":"higher","ROUGE-L":"higher","ROUGE (geometric mean of 1/2/L)":"higher"}},"counts":{"rows":10,"rows_with_code":0,"rows_with_paper_page":10,"rows_dated":10,"rows_using_additional_data":0},"rows":[{"rank_in_archive_order":1,"model":"NexusSum (Mistral Large)","metrics":{"BERTScore (F1)":"70.70","ROUGE (geometric mean of 1/2/L)":"18.27","ROUGE-1":"42.51","ROUGE-2":"10.27","ROUGE-L":"23.91"},"uses_additional_data":false,"paper_date":"2025-05-30","paper":"/paper/nexussum-hierarchical-llm-agents-for-long","paper_url":"https://arxiv.org/abs/2505.24575v1","paper_title":"NexusSum: Hierarchical LLM Agents for Long-Form Narrative Summarization","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":2,"model":"CachED (BART Large)","metrics":{"BERTScore (F1)":"54.4"},"uses_additional_data":false,"paper_date":"2025-01-03","paper":"/paper/end-to-end-long-document-summarization-using","paper_url":"https://arxiv.org/abs/2501.01805v1","paper_title":"End-to-End Long Document Summarization using Gradient Caching","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":3,"model":"SLED (BART Large)","metrics":{"BERTScore (F1)":"52.4"},"uses_additional_data":false,"paper_date":"2025-01-03","paper":"/paper/end-to-end-long-document-summarization-using","paper_url":"https://arxiv.org/abs/2501.01805v1","paper_title":"End-to-End Long Document Summarization using Gradient Caching","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":4,"model":"Unlimiformer (BART Base)","metrics":{"BERTScore (F1)":"51.5"},"uses_additional_data":false,"paper_date":"2025-01-03","paper":"/paper/end-to-end-long-document-summarization-using","paper_url":"https://arxiv.org/abs/2501.01805v1","paper_title":"End-to-End Long Document Summarization using Gradient Caching","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":5,"model":"Zero-Shot (GPT-4o)","metrics":{"BERTScore (F1)":"47.24","ROUGE-1":"20.3","ROUGE-2":"3.5","ROUGE-L":"17.68"},"uses_additional_data":false,"paper_date":"2025-01-03","paper":"/paper/end-to-end-long-document-summarization-using","paper_url":"https://arxiv.org/abs/2501.01805v1","paper_title":"End-to-End Long Document Summarization using Gradient Caching","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":6,"model":"Zero-Shot (Mistral Large)","metrics":{"BERTScore (F1)":"46.42","ROUGE-1":"19.63","ROUGE-2":"2.99","ROUGE-L":"12.0"},"uses_additional_data":false,"paper_date":"2025-05-30","paper":"/paper/nexussum-hierarchical-llm-agents-for-long","paper_url":"https://arxiv.org/abs/2505.24575v1","paper_title":"NexusSum: Hierarchical LLM Agents for Long-Form Narrative Summarization","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":7,"model":"Chain of Agents (Claude 3 Opus)","metrics":{"ROUGE (geometric mean of 1/2/L)":"17.47"},"uses_additional_data":false,"paper_date":"2024-06-04","paper":"/paper/chain-of-agents-large-language-models","paper_url":"https://arxiv.org/abs/2406.02818v1","paper_title":"Chain of Agents: Large Language Models Collaborating on Long-Context Tasks","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":8,"model":"NexusSum (Claude 3 Haiku)","metrics":{"ROUGE (geometric mean of 1/2/L)":"16.46"},"uses_additional_data":false,"paper_date":"2025-05-30","paper":"/paper/nexussum-hierarchical-llm-agents-for-long","paper_url":"https://arxiv.org/abs/2505.24575v1","paper_title":"NexusSum: Hierarchical LLM Agents for Long-Form Narrative Summarization","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":9,"model":"Chain of Agents (Claude 3 Sonet)","metrics":{"ROUGE (geometric mean of 1/2/L)":"14.96"},"uses_additional_data":false,"paper_date":"2024-06-04","paper":"/paper/chain-of-agents-large-language-models","paper_url":"https://arxiv.org/abs/2406.02818v1","paper_title":"Chain of Agents: Large Language Models Collaborating on Long-Context Tasks","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":10,"model":"Chain of Agents (Claude 3 Haiku)","metrics":{"ROUGE (geometric mean of 1/2/L)":"13.70"},"uses_additional_data":false,"paper_date":"2024-06-04","paper":"/paper/chain-of-agents-large-language-models","paper_url":"https://arxiv.org/abs/2406.02818v1","paper_title":"Chain of Agents: Large Language Models Collaborating on Long-Context Tasks","code":null,"n_code_links":0,"syntology":null}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,885 of the 9,623 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9623,"papers_checked":6885,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":2737},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-25T09:33:49+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":0,"rows_with_any_sample_ran":0,"distinct_papers_with_graph_line":0,"distinct_papers_with_any_sample_ran":0,"samples_over_distinct_papers":{"n_ran":0,"n_unverified":0,"n_samples":0,"n_pointer_only_licence":0,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":0,"n_unverified":0,"n_samples":0,"n_pointer_only_licence":0,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}