[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86139-en":3,"doc-seo-86139-105":29,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":11,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},86139,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Dynamic Graph Continual Learning via Condensation and Attachment","Dynamic graph continual learning (DGCL) addresses catastrophic forgetting as graphs evolve through changing nodes and edges. Existing DGCL approaches fail to fully leverage temporal information across graph snapshots. CA-DGCL proposes condensing historical snapshots into compact semantic representations, building cross-timestamp node chains to form a third-order tensor, and applying Tucker decomposition to obtain stable node features. These features generate and attach historically informed nodes to the current graph for replay, together with a refined forgetting measure. Extensive experiments show improved forgetting suppression and competitive accuracy.","CA-DGCL: Dynamic Graph Continual Learning via Condensation and  \nAttachment  \nTingxu Yan 1 and Ye Yuan*  \narXiv :2607 . 1 1 1 12v 1 [ cs .LG] 13 Jul 2026  \nAbstract—Dynamic graph continual learning (DGCL) isan effective manner for handling catastrophic forgetting in dynamic graphs. However, existing DGCL methods underutilize temporal information across graph snapshots. To address this critical issue, we propose a novel framework for Dynamic Graph Continual Learning via Condensation and Attachment (CA-DGCL). Specifically, CA-DGCL first condenses historical graph snapshots into compact semantic representations efficiently. Further, a cross-timestamp node chains is built to construct a third-order tensor and Tucker decomposition is applied to this tensor for obtaining stable node features, which encapsulate historical knowledge. Finally, these node features are used to generate new nodes and attached to the current graph for replaying of past information without compromising the new patterns. In addtion, a refined forgetting measure is introduced to make it more suitable for dynamic graph settings. Extensive experiments demonstrate that CA-DGCL outperforms baselines in forgetting suppression as well as maintain competitive accuracy, proving its efficacy for dynamic graph continual learning.  \nI. INTRODUCTION  \nGraphs serve as fundamental structures for modeling relational data across domains such as social networks. Graph Neural Networks (GNNs) have emerged as the predominant approach for graph representation learning, achieving considerable success. Recent graph-oriented studies have further explored dynamic graph modeling, graph-based clustering, neural community search, graph pooling, and graphregularized representation learning [31], [34], [37]–[39],[41],[51] . Meanwhile, representative GNN architectures and scalable training strategies, such as attention-based aggregation, expressive graph isomorphism networks, personalized propagation, simplified graph convolution, clustering-based mini-batching, sampling-based training, edge dropping, and self-supervised graph representation learning, have broadened the practical applicability of graph learning [77]–[84] . Nevertheless, real-world graphs are inherently dynamic, continuously evolving through the addition or removal of nodes and edges. This temporal aspect poses significant challenges for conventional GNNs tailored to static graphs, often resulting in catastrophic forgetting where newly learned patterns overwrite previously acquired knowledge.  \nAlthough continual learning research has been extended to dynamic graph settings, existing methods do not effectively exploit the temporal information inherent in sequences of graph snapshots to capture intrinsic node characteristics.  \n1Tingxu Yan and Ye Yuan* are with the College of Computer and Information Science, Southwest University, Chongqing 400715, China  \n[a020909@email.swu.edu.cn](a020909@email.swu.edu.cn) ,[yuanyekl@swu.edu.cn](yuanyekl@swu.edu.cn)  \nRelated studies on dynamic, spatio-temporal, and tensorbased representation learning demonstrate the importance of modeling temporal evolution and high-order dependencies in non-stationary data [27], [29], [32], [35], [36], [40], [42],[44], [53], [60],[62] . Dynamic graph representation learning methods further show that evolving node states, temporal self-attention, continuous-time interactions, temporal neighborhoods, and triadic closure are important for capturing non-stationary graph patterns [85]–[90] . Meanwhile, current evaluation frameworks also lack adaptations specific to the properties of dynamic graphs. Metrics borrowed from static domains—such as forgetting rate—measure a model’s performance on prior tasks after training on subsequent ones. However, in practical scenarios where tasks are segmented by time in dynamic graphs, retrospective performance on earlier tasks becomes less relevant. Taking node classification as an example, greater emphasis should instead b","cbCaikWY3l7URbbh","https://ap.wps.com/l/cbCaikWY3l7URbbh","pdf",1537032,3,1,"English","en",105,"# Introduction\n## Problem: catastrophic forgetting in dynamic graphs\n## Background and limitations of existing DGCL and evaluation\n## Proposed framework: CA-DGCL and main contributions","[{\"question\":\"What problem does CA-DGCL target in dynamic graph continual learning?\",\"answer\":\"It targets catastrophic forgetting caused by evolving graphs where new patterns can overwrite previously learned knowledge.\"},{\"question\":\"How does CA-DGCL use historical information from past graph snapshots?\",\"answer\":\"It condenses historical snapshots into compact semantic representations, then constructs cross-timestamp node chains to build a tensor and extract stable node features via Tucker decomposition.\"},{\"question\":\"How does CA-DGCL replay past knowledge while adapting to the current graph?\",\"answer\":\"It generates new nodes enriched with historical information from the stable features and selectively attaches them to the current task’s graph to preserve useful past knowledge.\"}]",1784208862,20,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"dynamic-graph-continual-learning-via-condensation-and-attachment","",{"@graph":35,"@context":84},[36,52,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,49],{"item":40,"name":41,"@type":42,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":20},"https://docshare.wps.com/document/research-report/",{"item":50,"name":13,"@type":42,"position":51},"https://docshare.wps.com/document/dynamic-graph-continual-learning-via-condensation-and-attachment/86139/",4,{"url":50,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":23,"description":14,"dateModified":60,"datePublished":61,"encodingFormat":59,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":40,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-27","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":20},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does CA-DGCL target in dynamic graph continual learning?","Question",{"text":74,"@type":75},"It targets catastrophic forgetting caused by evolving graphs where new patterns can overwrite previously learned knowledge.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does CA-DGCL use historical information from past graph snapshots?",{"text":79,"@type":75},"It condenses historical snapshots into compact semantic representations, then constructs cross-timestamp node chains to build a tensor and extract stable node features via Tucker decomposition.",{"name":81,"@type":72,"acceptedAnswer":82},"How does CA-DGCL replay past knowledge while adapting to the current graph?",{"text":83,"@type":75},"It generates new nodes enriched with historical information from the stable features and selectively attaches them to the current task’s graph to preserve useful past knowledge.","https://schema.org",{"og:url":50,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":50},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,126,129,133],{"id":21,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":51,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":28,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":28,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":28,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]