面向知識圖譜的網(wǎng)絡(luò)信息自監(jiān)督強化學(xué)習(xí)推薦模型

打開文本圖片集

引用格式：.面向知識圖譜的網(wǎng)絡(luò)信息自監(jiān)督強化學(xué)習(xí)推薦模型[J].現(xiàn)代電子技術(shù)，2025，48（10）：142-146

關(guān)鍵詞：知識圖譜；網(wǎng)絡(luò)信息；自監(jiān)督；強化學(xué)習(xí)；推薦模型；交互信息；特征提?。幌嗨贫扔嬎阒袌D分類號：TN711-34；TP182 文獻(xiàn)標(biāo)識碼：A 文章編號：1004-373X（2025）10-0142-05

Abstract：Inorder todeeplyunderstandandmine thebehavioralcharacteristicsof user historicalnetworkinteraction information，dynamicallextractchangesinuserinteractionbehaviorandachievepersonalizedrecommendationofnetwork information，aknowledgegraph basednetwork informationself-supervisedreinforcementlearningrecommendationmodelis constructed.Inthemodel，aknowledgegraphofusernetworkinformationinteractionbehaviorisonstructedtoclearlydisplay user'shistoricalnetwork information interactionbehavior.Thedynamicchangesofuserbehaviorintheknowledgegraphcanbe capturedefectivelybymeansofthefeatureextractionmodelbasedonself-supervisedreinforcement learningtoavoidthe negativeimpactofpopularityias，soastoextractthefeaturesof historicalnetworkinteractioninformation.Basedonknowledge graphsimilaritycalculation，thenetworkinformationentitieswithsimilarfeatures touserhistorical interaction informationare recommendedtorealizetheaccurateandpersonalizedrecommendations.Theexperimentalresultsverifiesthatafter recommending online movie information resources to users，the click play conversion rate can reach 96.83% ，and the personalized recommendation effect of online information is improved significantly.

Keywords：knowledgegraph；network information；self supervision；reinforcement learning；recommendationmodel; interactive information; feature extraction; similarity calculation

0 引言

隨著社交媒體、電子商務(wù)和在線視頻等領(lǐng)域的繁榮發(fā)展，網(wǎng)絡(luò)信息呈現(xiàn)出多樣化、海量化和快速更新的特點，這使得用戶難以從海量信息中篩選出符合自己興趣和需求的內(nèi)容。（剩余5834字）