[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-515283-105":59,"doc-detail-515283-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","reward-based-input-construction-for-cross-document-relation-extraction-515283","Reward-based Input Construction for Cross-document Relation Extraction","","Relation extraction (RE) identifies relationships between target entities in text, traditionally focusing on single sentences or single documents. Cross-document RE extends this to long documents where entity pairs may span multiple texts, but document-level inputs are limited by pretrained language model token constraints. This work proposes REward-based Input Construction (REIC), a learning-based sentence selector that retrieves relational evidence and trains it with reinforcement learning using RE prediction scores as rewards. Experiments show improved performance over heuristic baselines across structures and backbones.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/reward-based-input-construction-for-cross-document-relation-extraction-515283/515283/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/reward-based-input-construction-for-cross-document-relation-extraction-515283/515283.png","ImageObject",300,407,{"name":92,"@type":93},"Mimi","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-15","2026-10-06",true,{"@type":102,"interactionType":103,"userInteractionCount":52},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does cross-document relation extraction address?","Question",{"text":112,"@type":113},"It extracts relational facts when the target entity pair may not appear within the same sentence or even the same document, requiring reasoning across multiple long documents.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"Why is sentence selection important for long-document cross-document RE?",{"text":117,"@type":113},"Long documents exceed typical token limits of pretrained language models, so selecting informative sentences is necessary to preserve relational evidence while staying within input constraints.",{"name":119,"@type":110,"acceptedAnswer":120},"How does REIC train the sentence selection module without evidence supervision?",{"text":121,"@type":113},"REIC uses reinforcement learning, treating the relation prediction score from the selected sentences as the reward to iteratively improve selection.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},515283,1791428654,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":52,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145,"preview_page_count":19,"preview_urls":146},2336477974920,"https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d","Reward-based Input Construction for Cross-document Relation Extraction  \nByeonghu Na 1 * , Suhyeon Jo 1 * , Yeongmin Kim 1 , Il-Chul Moon 1 2  \n1 KAIST, [2](2 summary.ai)[ summary.ai](2 summary.ai)  \n{[byeonghu.na](byeonghu.na),suhyeonjo,alsdudrla10, [icmoon}@kaist.ac.kr](icmoon}@kaist.ac.kr)  \nAbstract  \nRelation extraction (RE) is a fundamental task in natural language processing, aiming to identify relations between target entities in text.  \nWhile many RE methods are designed for a single sentence or document, cross-document RE has emerged to address relations across multiple long documents. Given the nature of long documents in cross-document RE, extracting document embeddings is challenging due to the length constraints of pre-trained language models. Therefore, we propose REward-based Input Construction (REIC), the first learningbased sentence selector for cross-document RE.  \nREIC extracts sentences based on relational evidence, enabling the RE module to effectively infer relations. Since supervision of evidence sentences is generally unavailable, we train REIC using reinforcement learning with RE prediction scores as rewards. Experimental results demonstrate the superiority of our method over heuristic methods for different RE structures and backbones in crossdocument RE. Our code is publicly available at [https://github.com/aailabkaist/REIC](https://github.com/aailabkaist/REIC).  \n1 Introduction  \nThe task of relation extraction (RE) aims to identify relations between a pair of target entities in a given text (Zeng et al., 2014 ; Zhang et al., 2017 ; Sahu et al., 2019 ; Xu et al., 2021) . This task is fundamental in natural language processing (NLP) and provides crucial information for applications such as information retrieval (Kadry and Dietz, 2017) and question answering (Li et al., 2019 ; Chen et al., 2019) . Most existing RE methods are limited to scenarios where the entity pair is within a single sentence (Wang et al., 2016 ; Zhu et al., 2019) ora single document (Nan et al., 2020 ; Huang et al., 2021 ; Xu et al., 2021) . However, many scenarios require extracting relations across multiple documents, where the target entity pair may not coexist  \n*  \nEqual contribution  \nwithin the same document. Therefore, recent efforts to address these challenges have led to the proposal of cross-document RE by Yao et al. (2021), and research in this area has received considerable attention (Wang et al., 2022 ; Wu et al., 2023 ; Lu et al., 2023 ; Son et al., 2023) .  \nUnlike traditional RE research, cross-document RE involves extracting relational facts from largescale long documents. For example, the DocRED dataset (Yao et al., 2019), developed for documentlevel RE, contains an average of 198 words per document. In contrast, the CodRED dataset (Yao et al., 2021), designed for cross-document RE, hasan average of 2,416 words per document. Given the substantial length of documents in cross-document RE tasks, extracting document embeddings from pre-trained language models, a step common to all methods, poses considerable challenges. This is because pre-trained language models are limited by the maximum number of tokens they can process; for example, BERT (Devlin et al., 2019) typically has a limit of 512 tokens.  \nA simple approach adopted in several studies is to use text snippets around the target entities as input to pre-trained language models, as shown in the dashed boxes in the middle panel of Figure 1. Often, this text snippet is determined by either word proximity or document structure without any adjustment. However, if only a subset of sentences around the entities is extracted from long documents, there is a risk of missing important sentences that are crucial for RE. For example, as seen in the left panel of Figure 1, using snippets around the entities fails to capture the relation ‘platform’between the entity pair (Kubuntu, x86-64) .  \nIn this paper, we introduce REward-based Input Construction (REIC), the first le","cbCaitrdwQtbuPcZ","https://ap.wps.com/l/cbCaitrdwQtbuPcZ","pdf",782986,17,"English","# Abstract\n# 1 Introduction","[{\"question\":\"What problem does cross-document relation extraction address?\",\"answer\":\"It extracts relational facts when the target entity pair may not appear within the same sentence or even the same document, requiring reasoning across multiple long documents.\"},{\"question\":\"Why is sentence selection important for long-document cross-document RE?\",\"answer\":\"Long documents exceed typical token limits of pretrained language models, so selecting informative sentences is necessary to preserve relational evidence while staying within input constraints.\"},{\"question\":\"How does REIC train the sentence selection module without evidence supervision?\",\"answer\":\"REIC uses reinforcement learning, treating the relation prediction score from the selected sentences as the reward to iteratively improve selection.\"}]","Reward-based Input Construction for Cross-document Relation Extraction | PDF",1791311307,43,[147,151,155,159],{"page_no":8,"url":148,"width":149,"height":150},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/9fc39954c33064fac02f72be4d0f8cf8.webp",983,1390,{"page_no":14,"url":152,"width":153,"height":154},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/d482e0d2391c87120f5e5503d150d5f9.webp",982,1388,{"page_no":81,"url":156,"width":157,"height":158},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/1f9a9725ba7fb149f0149dca540970e3.webp",981,1387,{"page_no":19,"url":160,"width":161,"height":162},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/386a2471e60b5113f9de9ca748992b5b.webp",980,1385]