{"id":1353,"date":"2018-08-01T15:45:21","date_gmt":"2018-08-01T15:45:21","guid":{"rendered":"http:\/\/cricca.disi.unitn.it\/montresor\/?page_id=1353"},"modified":"2025-08-20T20:19:52","modified_gmt":"2025-08-20T20:19:52","slug":"network-representation-learning","status":"publish","type":"page","link":"http:\/\/cricca.disi.unitn.it\/montresor\/research\/topics\/network-representation-learning\/","title":{"rendered":"Network representation learning"},"content":{"rendered":"

Between 2016 and 2019, thanks to the efforts of two PhD stu\u00adden\u00adts under my super\u00advi\u00adsion, Zekarias Kefato and Nasrullah Sheikh, I began explo\u00adring pro\u00adblems com\u00adple\u00adte\u00adly outsi\u00adde the distri\u00adbu\u00adted systems field\u2014namely, machi\u00adne lear\u00adning, with a par\u00adti\u00adcu\u00adlar focus on net\u00adwork repre\u00adsen\u00adta\u00adtion lear\u00adning. Perhaps it is my aca\u00adde\u00admic age, but when I say \u201cI began explo\u00adring,\u201d what I real\u00adly mean is that Zekarias and Nasrullah did 95% of the work, whi\u00adle my role was main\u00adly to ensu\u00adre that the papers were easy to read.<\/p>\n

Network Representation Learning (NRL<\/span>) is a method for lear\u00adning a low-dimen\u00adsio\u00adnal embed\u00adding of a gra\u00adph in such a way that its geo\u00adme\u00adtric pro\u00adper\u00adties are pre\u00adser\u00adved. The lear\u00adned embed\u00addings can then be applied to various down\u00adstream machi\u00adne lear\u00adning tasks, such as clas\u00adsi\u00adfi\u00adca\u00adtion and link prediction.<\/p>\n

NRL<\/span> can leve\u00adra\u00adge dif\u00adfe\u00adrent sour\u00adces of infor\u00adma\u00adtion in a gra\u00adph, such as net\u00adwork struc\u00adtu\u00adre, attri\u00adbu\u00adtes, and casca\u00addes. These sour\u00adces may be used inde\u00adpen\u00adden\u00adtly or in com\u00adbi\u00adna\u00adtion, depen\u00adding on their avai\u00adla\u00adbi\u00adli\u00adty. Early research relied pri\u00adma\u00adri\u00adly on struc\u00adtu\u00adral infor\u00adma\u00adtion becau\u00adse it is always avai\u00adla\u00adble by default. More recent work sug\u00adgests that incor\u00adpo\u00adra\u00adting addi\u00adtio\u00adnal infor\u00adma\u00adtion can lead to bet\u00adter repre\u00adsen\u00adta\u00adtions. The main chal\u00adlen\u00adge lies in how to effec\u00adti\u00adve\u00adly inte\u00adgra\u00adte dif\u00adfe\u00adrent sour\u00adces of infor\u00adma\u00adtion into the lear\u00adning process.<\/p>\n

Towards this end, we wor\u00adked on two directions: <\/span><\/p>\n

    \n
  1. Network repre\u00adsen\u00adta\u00adtion lear\u00adning on attri\u00adbu\u00adted gra\u00adphs <\/span>and hete\u00adro\u00adge\u00adneous gra\u00adphs<\/span>. <\/span>GAT2VEC<\/span> [Computing19]<\/a><\/span> <\/b>learns a repre\u00adsen\u00adta\u00adtion of nodes from struc\u00adtu\u00adral con\u00adtext and attri\u00adbu\u00adte con\u00adtext obtai\u00adned from struc\u00adtu\u00adral and attri\u00adbu\u00adte infor\u00adma\u00adtion respec\u00adti\u00adve\u00adly. The <\/span>HETNET2VEC<\/span> [SNAMS18b<\/a>]<\/span><\/b> model is for hete\u00adro\u00adge\u00adnous net\u00adwork repre\u00adsen\u00adta\u00adtion lear\u00adning. The model pre\u00adser\u00adves the various seman\u00adtic rela\u00adtion\u00adship among nodes to learn a representation. <\/span><\/li>\n
  2. Using casca\u00adde infor\u00adma\u00adtion for net\u00adwork repre\u00adsen\u00adta\u00adtion lear\u00adning [MLG17<\/span><\/a>]<\/span> [MLG17<\/span><\/a>][LOD18<\/span><\/a>] and vira\u00adli\u00adty pre\u00addic\u00adtion [SNAMS18a<\/a>]. <\/span>In the case of social net\u00adworks, the under\u00adly\u00ading net\u00adwork infor\u00adma\u00adtion may not be avai\u00adla\u00adble due to pro\u00advi\u00adder restric\u00adtions, but we can obser\u00adve the dif\u00adfu\u00adsion even\u00adts which are signals of the under\u00adly\u00ading net\u00adworks. Using casca\u00addes we can use reco\u00adver the under\u00adly\u00ading net\u00adwork throu\u00adgh net\u00adwork repre\u00adsen\u00adta\u00adtion learning.<\/li>\n<\/ol>\n

    Additional resul\u00adts have been obtai\u00adned in the  field of influen\u00adcer detec\u00adtion [MAISON17<\/span>]<\/a>, net\u00adwork inference\/link pre\u00addic\u00adtion  [MOD17<\/span>]<\/a><\/p>\n

    \n

    [MAISON17<\/span>]   <\/span><\/span>Zekarias T. Kefato and Alberto Montresor. Personalized influen\u00adcer detec\u00adtion: Topic and expo\u00adsu\u00adre-con\u00adfor\u00admi\u00adty aware. In Proc. of the<\/span> International Workshop on Mining Actionable Insights from Social<\/span> Networks<\/span>, MAISoN\u201917. ACM<\/span>, February 2017. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [MLG17<\/span>]   <\/span><\/span>Zekarias T. Kefato, Nasrullah Sheikh, and Alberto Montresor. Deepinfer: Diffusion net\u00adwork infe\u00adren\u00adce throu\u00adgh repre\u00adsen\u00adta\u00adtion lear\u00adning. In Proc. of the 13th International Workshop on Mining and Learning With<\/span> Graphs<\/span>, MLG<\/span>\u201917. ACM<\/span>, August 2017. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [MOD17<\/span>]   <\/span><\/span>Zekarias T. Kefato, Nasrullah Sheikh, and Alberto Montresor. Mineral: Multi-modal net\u00adwork repre\u00adsen\u00adta\u00adtion lear\u00adning. In Proc. of the<\/span> 3rd International Conference on Machine Learning, Optimization and Big<\/span> Data<\/span>, MOD<\/span>\u201917. ACM<\/span>, September 2017. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [LOD18<\/span>]   <\/span><\/span>Zekarias T. Kefato, Nasrullah Sheikh, and Alberto Montresor. REFINE<\/span>: Representation lear\u00adning from dif\u00adfu\u00adsion even\u00adts. In Proc. of the<\/span> 4th Conference on Machine Learning, Optimization and Data scien\u00adce<\/span>, LOD<\/span>\u201918. Springer, September 2018. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [SNAMS18a]   <\/span><\/span>Zekarias T. Kefato, Nasrullah Sheikh, Leila Bahri, Amira Soliman, Alberto Montresor, and Sarunas Girdzijauskas. CAS2VEC<\/span>: net\u00adwork-agno\u00adstic casca\u00adde pre\u00addic\u00adtion in onli\u00adne social net\u00adworks. In Proc. of the 5th International Conference<\/span>on Social Networks Analysis, Management and Security (SNAMS<\/span> 2018)<\/span>, pages 72\u201379. IEEE<\/span>, October 2018. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [SNAMS18b]   <\/span><\/span>Nasrullah Sheikh, Zekarias T. Kefato, and Alberto Montresor. Semi-super\u00advi\u00adsed hete\u00adro\u00adge\u00adneous infor\u00adma\u00adtion net\u00adwork embed\u00adding for node clas\u00adsi\u00adfi\u00adca\u00adtion using 1D-CNN<\/span>. In Proc. of the 5th International Conference<\/span> on Social Networks Analysis, Management and Security (SNAMS<\/span> 2018)<\/span>, pages 177\u2013181. IEEE<\/span>, October 2018. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [Computing19]   Nasrullah Sheikh, Zekarias Kefato, and Alberto Montresor. GAT2VEC<\/span>: Representation lear\u00adning for attri\u00adbu\u00adted gra\u00adphs. Computing<\/em>, 101(3):187\u2013209, 2019. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n

    [CN19<\/span>]   Nasrullah Sheikh, Zekarias T. Kefato, and Alberto Montresor. A sim\u00adple approach to attri\u00adbu\u00adted gra\u00adph embed\u00adding via enhan\u00adced autoen\u00adco\u00adder. In Proceedings of the Eighth Int. Conference on Complex<\/em> Networks and Their Applications (COMPLEX<\/span> NETWORKS<\/span> 2019)<\/em>, volu\u00adme 881 of Studies in Computational Intelligence<\/em>, pages 797\u2013809. Springer, December 2019.  [PDF<\/span>]<\/a>, [Bibtex]<\/a><\/p>\n

    [WWW21<\/span>]   Zekarias T. Kefato, Sarunas Girdzijauskas, Nasrullah Sheikh, and Alberto Montresor. Dynamic embed\u00addings for inte\u00adrac\u00adtion pre\u00addic\u00adtion. In Jure Leskovec, Marko Grobelnik, Marc Najork, Jie Tang, and Leila Zia, edi\u00adtors, The Web Conference 2021<\/em>, WWW<\/span>\u201921, pages 1609\u20131618. ACM<\/span> \/ IW3C2<\/span>, April 2021. [PDF<\/span>]<\/a>, [Bibtex]<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"

    Between 2016 and 2019, thanks to the efforts of two PhD stu\u00adden\u00adts under my super\u00advi\u00adsion, Zekarias Kefato and Nasrullah Sheikh, I began explo\u00adring pro\u00adblems com\u00adple\u00adte\u00adly outsi\u00adde the distri\u00adbu\u00adted systems field\u2014namely, machi\u00adne lear\u00adning, with a par\u00adti\u00adcu\u00adlar focus on net\u00adwork repre\u00adsen\u00adta\u00adtion lear\u00adning. Perhaps it is my aca\u00adde\u00admic age, but when I say \u201cI began explo\u00adring,\u201d what I [\u2026]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":1394,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"wp_typography_post_enhancements_disabled":false,"footnotes":""},"class_list":["post-1353","page","type-page","status-publish","hentry","post"],"_links":{"self":[{"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/pages\/1353","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/comments?post=1353"}],"version-history":[{"count":9,"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/pages\/1353\/revisions"}],"predecessor-version":[{"id":5567,"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/pages\/1353\/revisions\/5567"}],"up":[{"embeddable":true,"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/pages\/1394"}],"wp:attachment":[{"href":"http:\/\/cricca.disi.unitn.it\/montresor\/wp-json\/wp\/v2\/media?parent=1353"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}