[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-273814-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-273814-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","mil-bert-classification-of-arbitrarily-large-text-with-performance-and-explanatory-guarantees","MIL-BERT - Classification of Arbitrarily Large Text with Performance and Explanatory Guarantees","","Many text classification and regression decisions can be supported by constituent excerpts rather than full documents. Inspired by multiple instance learning, MIL-BERT trains a neural model to classify text by selecting informative passages via fixed selectors. The approach scales to collections approaching 1M tokens and is evaluated on seven long-text datasets exceeding base-model encoding limits. State-of-the-art results are reported on political bias detection, trigger warnings, and author demographics, with generalization to smaller instances from weakly labeled bags.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/mil-bert-classification-of-arbitrarily-large-text-with-performance-and-explanatory-guarantees/273814/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/mil-bert-classification-of-arbitrarily-large-text-with-performance-and-explanatory-guarantees/273814.png","ImageObject",300,407,{"name":42,"@type":43},"Clementine","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-17","2026-09-15",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",8,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does MIL-BERT address in text classification?","Question",{"text":63,"@type":64},"It addresses the limitation of transformer models on long documents by classifying text using selected excerpts instead of processing entire overlong inputs.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does MIL-BERT apply multiple instance learning to text?",{"text":68,"@type":64},"It represents a long document as a bag of excerpt instances, where training uses bag-level labels and learns an instance classifier through selectors and pooling.",{"name":70,"@type":61,"acceptedAnswer":71},"What are the key benefits MIL-BERT claims beyond accuracy?",{"text":72,"@type":64},"MIL-BERT emphasizes explainability by requiring identification of key, easily reviewable passages that drive the classification decision.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},273814,1789472859,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":55,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":80,"read_time":144,"preview_page_count":4},1374391974564,"https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002","MIL-BERT: Classification of Arbitrarily Large Text with Performance and  \nExplanatory Guarantees  \nJohn Cadigan  \nSRI International  \n[john.cadigan@sri.com](john.cadigan@sri.com)  \nDayne Freitag  \n[dayne.freitag@sri.com](dayne.freitag@sri.com)  \nEric Yeh  \nSRI International [eric.yeh@sri.com](eric.yeh@sri.com)  \narXiv : 2608 .20636v 1 [cs .CL] 2 1 Aug 2026  \nAbstract  \nMany text classification decisions are viable based on constituent excerpts alone. Taking inspiration from the field of multiple instance learning, we present an algorithm for training a neural network to classify text by selecting such excerpts. We show that our approach is also scalable with demonstrated learning against samples with nearly 1M tokens. We evaluate our methods on 7 datasets with emphasis on long-textual collections that far exceed the encoding limit of our base model. We present state-of-the-art results with this algorithm on 3 datasets: identification of political bias in news outlets, trigger warnings in long stories, and demographic characteristics of authors in tweet collections. Furthermore, the model trained on weakly-labeled collections of text (bags) generalizes to accurately classify constituent, smaller instances. Besides a new state-of-the-art for these problems, this approach is one of the few neural methods to excel in these datasets.  \n1 Introduction  \nIn the current state of the art, the most accurate approaches for textual classification and regression (C&R) are based on language models in the transformer family (Vaswani et al., 2017) . However, canonical transformers have fixed-length positional embeddings and computation and memory usage that grows quadratically with sequence length. Many practical applications of C&R using popular foundation models make compromises such as truncation when processing long documents, implicitly assuming that sufficient information for accurate prediction can be found early in a document. On the other hand, transformer adaptations for long documents such as Infinite Transformer (Martinset al., 2022) are relatively expensive to apply and process the entire text, opaquely and possibly needlessly. We surmise that many problems of textual C&R can be successfully addressed by instead  \nidentifying one or a few passages that support a particular prediction—a phenomenon we call signal locality. For problems exhibiting signal locality, we can replace the wholesale processing of an overlarge text with a search for indicative excerpts of modest size. This amounts to a multiple instance learning (MIL) framing of the problem.  \nMIL is a paradigm of machine learning in which collections of instances called bags. A bag is classified as positive as long as one of its instances is positive. For example, MIL has been applied to the task of identifying cancerous image patches (instances) within larger, whole-slide images based on a slide-level (bag) label (Qu et al., 2024) . If animage contains cancer, then there is at least one image patch that contains cancer. If an image does not contain cancer, then no image patches contain cancer. An example application of MIL to text involves identifying civil unrest based on classifying collections of tweets for particular days and countries (DeLucia et al., 2023) . Inspired by these approaches, we develop an algorithm for learning to classify arbitrarily long text with performance and explainability guarantees.  \nThis paper makes the following contributions:  \n• We describe an efficient approach for classifying long texts based on a MIL framing, involving a fixed set of “selectors” that identify the passages most relevant to a particular prediction.  \n• In experiments involving a range of previously proposed long-text classification problems, we demonstrate that this excerptive approach, MIL-BERT, is competitive with whole-document alternatives, in several cases surpassing previously reported accuracies.  \n• We describe the advantages of MIL-BERT for explainability, arisin","cbCaijwfmlHoCjHN","https://ap.wps.com/l/cbCaijwfmlHoCjHN","pdf",1544079,11,"English","# Abstract\n# 1 Introduction\n## Signal locality and MIL framing\n## Contributions\n# 2 Method\n## 2.1 Preliminaries\n## 2.2 Algorithm","[{\"question\":\"What problem does MIL-BERT address in text classification?\",\"answer\":\"It addresses the limitation of transformer models on long documents by classifying text using selected excerpts instead of processing entire overlong inputs.\"},{\"question\":\"How does MIL-BERT apply multiple instance learning to text?\",\"answer\":\"It represents a long document as a bag of excerpt instances, where training uses bag-level labels and learns an instance classifier through selectors and pooling.\"},{\"question\":\"What are the key benefits MIL-BERT claims beyond accuracy?\",\"answer\":\"MIL-BERT emphasizes explainability by requiring identification of key, easily reviewable passages that drive the classification decision.\"}]","MIL-BERT - Classification of Arbitrarily Large Text with Performance and Explanatory Guarantees | PDF",28]