{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,8,1]],"date-time":"2025-08-01T06:10:04Z","timestamp":1754028604162,"version":"3.41.2"},"reference-count":32,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2025,8,1]],"date-time":"2025-08-01T00:00:00Z","timestamp":1754006400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:sec><jats:title>Introduction<\/jats:title><jats:p>Text summarization is a longstanding challenge in natural language processing, with recent advancements driven by the adoption of Large Language Models (LLMs) and Small Language Models (SLMs). Despite these developments, issues such as the \u201cLost in the Middle\u201d problem\u2014where LLMs tend to overlook information in the middle of lengthy prompts\u2014persist. Traditional summarization, often termed the \u201cStuff\u201d method, processes an entire text in a single pass. In contrast, the \u201cMap\u201d method divides the text into segments, summarizes each independently, and then synthesizes these partial summaries into a final output, potentially mitigating the \u201cLost in the Middle\u201d issue. This study investigates whether the Map method outperforms the Stuff method for texts that fit within the context window of SLMs and assesses its effectiveness in addressing the \u201cLost in the Middle\u201d problem.<\/jats:p><\/jats:sec><jats:sec><jats:title>Methods<\/jats:title><jats:p>We conducted a two-part investigation: first, a simulation study using generated texts, paired with an automated fact-retrieval evaluation to eliminate the need for human assessment; second, a practical study summarizing scientific papers.<\/jats:p><\/jats:sec><jats:sec><jats:title>Results<\/jats:title><jats:p>Results from both studies demonstrate that the Map method produces summaries that are at least as accurate as those from the Stuff method. Notably, the Map method excels at retaining key facts from the beginning and middle of texts, unlike the Stuff method, suggesting its superiority for SLM-based summarization of smaller texts. Additionally, SLMs using the Map method achieved performance comparable to LLMs using the Stuff method, highlighting its practical utility.<\/jats:p><\/jats:sec><jats:sec><jats:title>Discussion<\/jats:title><jats:p>Both theoretical and practical studies suggest that using Map method for summarization with SLM allowed to address the \u201cLost in the Middle\u201d problem and outperform Stuff method.<\/jats:p><\/jats:sec>","DOI":"10.3389\/frai.2025.1604034","type":"journal-article","created":{"date-parts":[[2025,8,1]],"date-time":"2025-08-01T05:32:16Z","timestamp":1754026336000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Divide and summarize: improve SLM text summarization"],"prefix":"10.3389","volume":"8","author":[{"given":"Alexandre","family":"Bailly","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Antoine","family":"Saubin","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Gabriel","family":"Kocevar","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Jonathan","family":"Bodin","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"1965","published-online":{"date-parts":[[2025,8,1]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2004.05150","article-title":"Longformer: the long-document transformer","author":"Beltagy","year":"2020","journal-title":"arXiv"},{"key":"B2","first-page":"1877","article-title":"\u201cLanguage models are few-shot learners,\u201d","author":"Brown","year":"2020","journal-title":"Advances in Neural Information Processing Systems, volume 33"},{"key":"B3","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2310.00785","article-title":"Booookscore: a systematic exploration of book-length summarization in the era of llms","author":"Chang","year":"2023","journal-title":"arXiv"},{"key":"B4","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1810.04805","article-title":"Bert: Pre-training of deep bidirectional transformers for language understanding","author":"Devlin","year":"2018","journal-title":"arXiv"},{"key":"B5","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2407.21783","article-title":"The llama 3 herd of models","author":"Dubey","year":"2024","journal-title":"arXiv"},{"key":"B6","unstructured":"Frontiers in Artificial Intelligence"},{"key":"B7","unstructured":"Frontiers in Astronomy and Space Sicences"},{"key":"B8","unstructured":"Frontiers in Bioengineering and Biotechnology"},{"key":"B9","unstructured":"Frontiers in Earth Science"},{"key":"B10","unstructured":"Frontiers in Education"},{"key":"B11","unstructured":"Frontiers in Neurosciences"},{"key":"B12","doi-asserted-by":"crossref","first-page":"123","DOI":"10.18653\/v1\/2024.nlp4pi-1.10","article-title":"\u201cEfficient aspect-based summarization of climate change reports with small language models,\u201d","volume-title":"Proceedings of the Third Workshop on NLP for Positive Impact","author":"Ghinassi","year":"2024"},{"key":"B13","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2310.06825","article-title":"Mistral 7b","author":"Jiang","year":"2023","journal-title":"arXiv"},{"key":"B14","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3545176","article-title":"An empirical survey on long document summarization: datasets, models, and metrics","volume":"55","author":"Koh","year":"2022","journal-title":"ACM Comput. Surv"},{"key":"B15","first-page":"74","article-title":"\u201cRouge: a package for automatic evaluation of summaries,\u201d","volume-title":"Text Summarization Branches Out","author":"Lin","year":"2004"},{"key":"B16","doi-asserted-by":"publisher","first-page":"157","DOI":"10.1162\/tacl_a_00638","article-title":"Lost in the middle: How language models use long contexts","volume":"12","author":"Liu","year":"2024","journal-title":"Trans. Assoc. Comput. Linguist"},{"key":"B17","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1911.02782","article-title":"S2ORC: the semantic scholar open research corpus","author":"Lo","year":"2019","journal-title":"arXiv"},{"key":"B18","doi-asserted-by":"publisher","first-page":"11085","DOI":"10.1609\/aaai.v36i10.21357","article-title":"Semantic self-segmentation for abstractive summarization of long documents in low-resource regimes","volume":"36","author":"Moro","year":"2022","journal-title":"Proc. AAAI Conf. Artif. Intell"},{"key":"B19","doi-asserted-by":"publisher","first-page":"3542","DOI":"10.3390\/s23073542","article-title":"Efficient memory-enhanced transformer for long-document summarization in low-resource regimes","volume":"23","author":"Moro","year":"2023","journal-title":"Sensors"},{"key":"B20","doi-asserted-by":"crossref","DOI":"10.1007\/978-1-4419-0318-1","volume-title":"Mixed-Effects Models in S and S-PLUS","author":"Pinheiro","year":"2000"},{"key":"B21","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2309.09558","article-title":"Summarization is (almost) dead","author":"Pu","year":"2023","journal-title":"arXiv"},{"key":"B22","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2408.00118","article-title":"Gemma 2: Improving open language models at a practical size","author":"Riviere","year":"2024","journal-title":"arXiv"},{"key":"B23","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3485766","article-title":"A survey of evaluation metrics used for NLG systems","volume":"55","author":"Sai","year":"2022","journal-title":"ACM Comput. Surv"},{"key":"B24","doi-asserted-by":"publisher","first-page":"7782","DOI":"10.3390\/app14177782","article-title":"Putting GPT-4O to the sword: a comprehensive evaluation of language, vision, speech, and multimodal proficiency","volume":"14","author":"Shahriar","year":"2024","journal-title":"Appl. Sci"},{"key":"B25","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1706.03762","article-title":"Attention is all you need","author":"Vaswani","year":"2017","journal-title":"arXiv"},{"key":"B26","doi-asserted-by":"publisher","first-page":"132","DOI":"10.23919\/JSC.2024.0010","article-title":"Segmented summarization and refinement: a pipeline for long-document analysis on social media","volume":"5","author":"Wang","year":"2024","journal-title":"J. Soc. Comput"},{"key":"B27","doi-asserted-by":"publisher","first-page":"421","DOI":"10.3390\/info11090421","article-title":"Measurement of text similarity: a survey","volume":"11","author":"Wang","year":"2020","journal-title":"Information"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2109.10862","article-title":"Recursively summarizing books with human feedback","author":"Wu","year":"2021","journal-title":"arXiv"},{"key":"B29","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2410.13155","article-title":"SLM-MOD: small language models surpass llms at content moderation","author":"Zhan","year":"2024","journal-title":"arXiv"},{"key":"B30","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2406.11289","article-title":"A systematic survey of text summarization: From statistical methods to large language models","author":"Zhang","year":"2024","journal-title":"arXiv"},{"key":"B31","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1904.09675","article-title":"Bertscore: Evaluating text generation with bert","author":"Zhang","year":"2019","journal-title":"arXiv"},{"key":"B32","doi-asserted-by":"crossref","first-page":"1592","DOI":"10.18653\/v1\/2022.acl-long.112","article-title":"\u201cSummn: a multi-stage summarization framework for long input dialogues and documents: a multi-stage summarization framework for long input dialogues and documents,\u201d","volume-title":"Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","author":"Zhang","year":"2022"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2025.1604034\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,8,1]],"date-time":"2025-08-01T05:32:20Z","timestamp":1754026340000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2025.1604034\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,8,1]]},"references-count":32,"alternative-id":["10.3389\/frai.2025.1604034"],"URL":"https:\/\/doi.org\/10.3389\/frai.2025.1604034","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,8,1]]},"article-number":"1604034"}}