[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-detail-579083-en":59,"doc-seo-579083-105":80},{"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":5,"data":60},{"doc_id":61,"user_id":62,"nickname":63,"user_avatar":64,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":66,"doc_content":67,"file_id":68,"file_url":69,"file_type":70,"file_size":71,"view_count":24,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":72,"language":73,"language_code":74,"site_id":75,"html_lang":74,"table_of_contents":76,"faqs":77,"seo_title":78,"seo_description":66,"update_tm":79,"read_time":26,"preview_page_count":4},579083,8796095462418,"Noah","https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780","Principled Transformers for Predictive Performance in Knowledge Tracing","Knowledge tracing models students’ evolving knowledge over time, supporting accurate predictions that drive adaptive sequencing and personalized feedback in intelligent tutoring systems. This paper introduces knowledge tracing set transformers (KTSTs), a straightforward transformer-based model class using a set-based representation of features per student interaction. KTSTs add MHSA aggregation and learnable ALiBi to simplify attention while enabling arbitrary extra interaction features. Experiments on real-world data and simulated sequences validate design choices and report new state-of-the-art results on benchmark datasets.","Principled Transformers for Predictive Performance in Knowledge Tracing  \nKai Neubauer∗ Leuphana University L neburg, Germany [kai.neubauer@leuphana.de](kai.neubauer@leuphana.de)  \nYannick Rudolph∗ Leuphana University L neburg, Germany [yannick.rudolph@leuphana.de](yannick.rudolph@leuphana.de)  \nUlf Brefeld Leuphana University L neburg, Germany [ulf.brefeld@leuphana.de](ulf.brefeld@leuphana.de)  \nKnowledge tracing aims to model students’ knowledge and abilities over time, which is crucial for intelligent tutoring systems. In this paper, we propose a straightforward model class, knowledge tracing set transformers (KTSTs), specifically addressing predictive performance in knowledge tracing tasks. KTSTs closely follow prominent transformer architectures and use an intuitive set-based representation for student interactions. We introduce learnable ALiBi, which simplifies and improves upon a prevalent attention mechanism in knowledge tracing, and MHSA aggregation, which readily allows incorporating an arbitrary number of additional, potentially more complex features per student interaction. We highlight and discuss flaws present in related approaches, which are overly complex and, in part, based on suboptimal design choices. We validate our design choices for KTSTs in experiments with real-world data and simulated learning sequences. Overall, we address lessons learned and propose a straightforward model that relies on best practices and establishes a new state-of-the-art on standardized benchmark datasets. Ultimately, KTSTs may serve as a simple but effective base model class for future research in knowledge tracing and intelligent tutoring systems. Code is available [at](at github.com/kainbr/kt)[ github.com/kainbr/kt](at github.com/kainbr/kt) set transformers.  \nKeywords: intelligent tutoring systems, knowledge tracing, deep learning, transformer  \n1. INTRODUCTION  \nModeling students is considered key for computer-aided personalized education (Self, 1974) . Knowledge tracing provides a principled way to model and reason about changes in students’knowledge and abilities and thus enables adaptive sequencing of learning materials and personalized feedback in intelligent tutoring systems. However, knowledge tracing is only useful if it accurately predicts students’ performance (Corbett and Anderson, 1994) . While reasoning about knowledge states can be achieved by post-hoc model-agnostic interpretability methods (e.g., Rodrigues et al. 2022), predictive performance of knowledge tracing models is imperative for informed instructional policies (cf. Gervet et al. 2020) . Accurate knowledge tracing models are thus essential for the overall performance of intelligent tutoring systems. In this paper, we focus on predictive performance and study knowledge tracing as a supervised sequential machine learning task in which we aim to predict the correctness of the next response of a student, given the history of their interactions with an intelligent tutoring system.  \n*Authors contributed equally to this work.  \n89 Journal of Educational Data Mining, Volume 18, No 1, 2026  \nTo achieve good predictive performance in knowledge tracing tasks, existing approaches often rely on domain knowledge that is specific to the educational learning setting, either regarding the representation of student interactions (Ghosh et al., 2020 ; Shen et al., 2022 ; Yinet al., 2023) or regarding model architectures (Zhang et al., 2017; Nakagawa et al., 2019; Long et al., 2021 ; Shen et al., 2021) . This results in rather complex model designs that are either very specific to certain dataset characteristics and features, which do not allow generalization to other knowledge tracing datasets, or that involve custom model components, which deviate from more generic machine learning approaches, making proper attribution of model performance difficult.1 Meanwhile, simpler models (e.g., Piech et al. 2015 ; Liu et al. 2023) do not achieve state-of-the-art or comparable perfo","cbCaiu0eoAkRABxg","https://ap.wps.com/l/cbCaiu0eoAkRABxg","pdf",462237,24,"English","en",105,"# Introduction\n## Predictive performance in knowledge tracing\n## Limitations of existing approaches\n# Proposed method: KTSTs\n## Set-based interaction representation\n## MHSA aggregation and learnable ALiBi\n## Incorporating additional features\n# Experimental validation and results","[{\"question\":\"What problem does the paper address in knowledge tracing?\",\"answer\":\"It targets predictive performance, aiming to forecast whether a student’s next response will be correct based on their interaction history in an intelligent tutoring system.\"},{\"question\":\"What are knowledge tracing set transformers (KTSTs)?\",\"answer\":\"KTSTs are a straightforward model class built around transformer architectures, using a set-based representation for the features associated with each student interaction.\"},{\"question\":\"Which model components does the paper propose to improve predictions?\",\"answer\":\"It proposes MHSA aggregation and learnable ALiBi, which improves and simplifies attention mechanisms for knowledge tracing while supporting additional interaction features.\"}]","Principled Transformers for Predictive Performance in Knowledge Tracing | PDF",1792034971,{"code":4,"msg":81,"data":82},"ok",{"site_id":75,"language":74,"slug":83,"title":65,"keywords":84,"description":66,"schema_data":85,"social_meta":139,"head_meta":141,"extra_data":143,"updated_unix":144},"principled-transformers-for-predictive-performance-in-knowledge-tracing","",{"@graph":86,"@context":138},[87,101,121],{"@type":88,"itemListElement":89},"BreadcrumbList",[90,94,96,99],{"item":91,"name":92,"@type":93,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":95,"name":9,"@type":93,"position":14},"https://docshare.wps.com/document/",{"item":97,"name":40,"@type":93,"position":98},"https://docshare.wps.com/document/research-report/",3,{"item":100,"name":65,"@type":93,"position":19},"https://docshare.wps.com/document/principled-transformers-for-predictive-performance-in-knowledge-tracing/579083/",{"url":100,"name":65,"@type":102,"image":103,"author":108,"headline":65,"publisher":110,"fileFormat":113,"inLanguage":74,"description":66,"dateModified":114,"datePublished":115,"encodingFormat":113,"isAccessibleForFree":116,"interactionStatistic":117},"DigitalDocument",{"url":104,"@type":105,"width":106,"height":107},"https://docshare.wps.com/thumbnails/principled-transformers-for-predictive-performance-in-knowledge-tracing/579083.png","ImageObject",300,407,{"name":63,"@type":109},"Person",{"url":91,"name":111,"@type":112},"DocShare","Organization","application/pdf","2026-11-16","2026-10-15",true,{"@type":118,"interactionType":119,"userInteractionCount":24},"InteractionCounter",{"@type":120},"ViewAction",{"@type":122,"mainEntity":123},"FAQPage",[124,130,134],{"name":125,"@type":126,"acceptedAnswer":127},"What problem does the paper address in knowledge tracing?","Question",{"text":128,"@type":129},"It targets predictive performance, aiming to forecast whether a student’s next response will be correct based on their interaction history in an intelligent tutoring system.","Answer",{"name":131,"@type":126,"acceptedAnswer":132},"What are knowledge tracing set transformers (KTSTs)?",{"text":133,"@type":129},"KTSTs are a straightforward model class built around transformer architectures, using a set-based representation for the features associated with each student interaction.",{"name":135,"@type":126,"acceptedAnswer":136},"Which model components does the paper propose to improve predictions?",{"text":137,"@type":129},"It proposes MHSA aggregation and learnable ALiBi, which improves and simplifies attention mechanisms for knowledge tracing while supporting additional interaction features.","https://schema.org",{"og:url":100,"og:type":140,"og:title":65,"og:site_name":111,"og:description":66},"article",{"robots":142,"canonical":100},"index,follow",{"doc_id":61,"site_id":75},1792464369]