{"id":2101,"date":"2017-01-27T06:56:31","date_gmt":"2017-01-27T06:56:31","guid":{"rendered":"http:\/\/docs.uxart.io\/financo\/docs\/faqs\/getting-started\/required-plugins\/"},"modified":"2025-12-15T15:15:20","modified_gmt":"2025-12-15T15:15:20","slug":"required-plugins","status":"publish","type":"docs","link":"https:\/\/sahelib.atatec-design.com\/index.php\/docs\/gullu-knowledge-base\/getting-started\/required-plugins\/","title":{"rendered":"Syst\u00e8mes Recommandation Bas\u00e9s IA pour Plateformes Acad\u00e9miques"},"content":{"rendered":"<h3><strong data-start=\"69\" data-end=\"90\">R\u00e9sum\u00e9 (fran\u00e7ais)<\/strong><\/h3>\n<p>La croissance explosive de la production scientifique rend la d\u00e9couverte d\u2019articles pertinents de plus en plus difficile. Les syst\u00e8mes de recommandation \u2014 d\u00e9j\u00e0 efficaces dans le commerce et les m\u00e9dias \u2014 constituent une solution prometteuse pour les plateformes acad\u00e9miques. Cet article propose une synth\u00e8se compl\u00e8te : r\u00e9sum\u00e9, abstract, introduction, \u00e9tat de l\u2019art (revue syst\u00e9matique), analyse comparative des approches (contenu \/ collaboratif \/ hybride \/ deep learning), enjeux techniques (recommandation en temps r\u00e9el, architectures et technologies), recommandations pratiques et bibliographie s\u00e9lective. Il s\u2019appuie sur une revue des travaux et d\u2019un m\u00e9moire r\u00e9cent portant sur la conception d\u2019une plateforme de recommandation d\u2019articles scientifiques en temps r\u00e9el.<\/p>\n<h1 data-start=\"1572\" data-end=\"1588\">Introduction<\/h1>\n<p data-start=\"1589\" data-end=\"2391\">L\u2019acc\u00e8s \u00e0 la litt\u00e9rature scientifique est d\u00e9sormais confront\u00e9 \u00e0 un double d\u00e9fi : l\u2019abondance de publications (des milliers chaque jour) et la diversit\u00e9 des besoins des utilisateurs (chercheurs, doctorants, enseignants). Les moteurs de recherche traditionnellement bas\u00e9s sur mots-cl\u00e9s retournent trop souvent des listes volumineuses peu personnalis\u00e9es. Les syst\u00e8mes de recommandation appliqu\u00e9s au domaine acad\u00e9mique visent \u00e0 r\u00e9duire le temps de veille et am\u00e9liorer la pertinence des d\u00e9couvertes en combinant m\u00e9tadonn\u00e9es, texte int\u00e9gral et comportements d\u2019usage. Le pr\u00e9sent article expose les fondements th\u00e9oriques, compare les approches, d\u00e9taille la pratique des solutions temps r\u00e9el et propose des recommandations pour la mise en \u0153uvre sur plateformes acad\u00e9miques.<\/p>\n<h2 data-start=\"2446\" data-end=\"2473\">M\u00e9thodologie de la revue<\/h2>\n<p data-start=\"2474\" data-end=\"3155\">La litt\u00e9rature couvre trois grandes classes de m\u00e9thodes : filtrage bas\u00e9 sur le contenu, filtrage collaboratif et approches hybrides. De plus, depuis la derni\u00e8re d\u00e9cennie, les m\u00e9thodes profondes (embeddings, mod\u00e8les de langage comme BERT) et les architectures temps r\u00e9el (Kafka, Spark Streaming) \u00e9mergent pour r\u00e9pondre \u00e0 la dynamique des flux d\u2019articles et des interactions utilisateurs. Les travaux de synth\u00e8se du m\u00e9moire que nous exploitons recensent articles applicatifs (\u00e9ducation, e-sant\u00e9), prototypes acad\u00e9miques (ArZiGo) et plateformes (Google Scholar, ScienceDirect) pour \u00e9tablir forces\/faiblesses. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<h2 data-start=\"3157\" data-end=\"3190\">Filtrage bas\u00e9 sur le contenu<\/h2>\n<ul data-start=\"3191\" data-end=\"3733\">\n<li data-start=\"3191\" data-end=\"3357\">\n<p data-start=\"3193\" data-end=\"3357\"><strong data-start=\"3193\" data-end=\"3205\">Principe<\/strong> : repr\u00e9senter articles et profils via caract\u00e9ristiques textuelles (mots-cl\u00e9s, r\u00e9sum\u00e9s, sujets) puis recommander selon similarit\u00e9 (cosinus, distance).<\/p>\n<\/li>\n<li data-start=\"3358\" data-end=\"3428\">\n<p data-start=\"3360\" data-end=\"3428\"><strong data-start=\"3360\" data-end=\"3374\">Techniques<\/strong> : TF-IDF, word2vec, sentence embeddings, BERT, etc.<\/p>\n<\/li>\n<li data-start=\"3429\" data-end=\"3582\">\n<p data-start=\"3431\" data-end=\"3582\"><strong data-start=\"3431\" data-end=\"3444\">Avantages<\/strong> : fonctionne sans grande communaut\u00e9 d\u2019utilisateurs ; interpr\u00e9table ; \u00e9vite le cold-start pour les nouveaux items si m\u00e9tadonn\u00e9es riches.<\/p>\n<\/li>\n<li data-start=\"3583\" data-end=\"3733\">\n<p data-start=\"3585\" data-end=\"3733\"><strong data-start=\"3585\" data-end=\"3596\">Limites<\/strong> : risque de sur-sp\u00e9cialisation (peu de diversit\u00e9), d\u00e9pend fortement de la qualit\u00e9 des m\u00e9tadonn\u00e9es. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<p class=\"not-prose mt-0! mb-0! flex-auto truncate\">ETAT DE L&#8217;ART SUR LE SUJET DU\u2026<\/p>\n<\/li>\n<\/ul>\n<h2 data-start=\"3735\" data-end=\"3761\">Filtrage collaboratif<\/h2>\n<ul data-start=\"3762\" data-end=\"4354\">\n<li data-start=\"3762\" data-end=\"3950\">\n<p data-start=\"3764\" data-end=\"3950\"><strong data-start=\"3764\" data-end=\"3776\">Principe<\/strong> : exploiter comportements (clics, t\u00e9l\u00e9chargements, favoris, \u00e9valuations) pour rapprocher utilisateurs similaires ou items similaires \u00e0 partir de matrices utilisateur\u00d7item.<\/p>\n<\/li>\n<li data-start=\"3951\" data-end=\"4096\">\n<p data-start=\"3953\" data-end=\"4096\"><strong data-start=\"3953\" data-end=\"3966\">Variantes<\/strong> : memory-based (k-NN), model-based (matrix factorization, SVD), factorisation implicite (ALS), embeddings d\u2019utilisateurs\/items.<\/p>\n<\/li>\n<li data-start=\"4097\" data-end=\"4199\">\n<p data-start=\"4099\" data-end=\"4199\"><strong data-start=\"4099\" data-end=\"4112\">Avantages<\/strong> : capte pr\u00e9f\u00e9rences \u00e9mergentes ; bien adapt\u00e9 quand la base d\u2019utilisateurs est large.<\/p>\n<\/li>\n<li data-start=\"4200\" data-end=\"4354\">\n<p data-start=\"4202\" data-end=\"4354\"><strong data-start=\"4202\" data-end=\"4213\">Limites<\/strong> : cold-start (nouveaux utilisateurs\/items), questions de confidentialit\u00e9, performances en mise \u00e0 jour. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ul>\n<h2 data-start=\"4356\" data-end=\"4379\">Approches hybrides<\/h2>\n<ul data-start=\"4380\" data-end=\"4781\">\n<li data-start=\"4380\" data-end=\"4478\">\n<p data-start=\"4382\" data-end=\"4478\"><strong data-start=\"4382\" data-end=\"4394\">Principe<\/strong> : combiner contenu et collaboratif (pond\u00e9ration, cascade, commutation, stacking).<\/p>\n<\/li>\n<li data-start=\"4479\" data-end=\"4588\">\n<p data-start=\"4481\" data-end=\"4588\"><strong data-start=\"4481\" data-end=\"4492\">Int\u00e9r\u00eat<\/strong> : r\u00e9duit le probl\u00e8me de cold-start et b\u00e9n\u00e9ficie des forces compl\u00e9mentaires des deux familles.<\/p>\n<\/li>\n<li data-start=\"4589\" data-end=\"4781\">\n<p data-start=\"4591\" data-end=\"4781\"><strong data-start=\"4591\" data-end=\"4603\">Pratique<\/strong> : la plupart des plateformes acad\u00e9miques prototypes utilisent un hybride configurable (p. ex. combiner score contenu + score collaboratif). <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ul>\n<h2 data-start=\"4783\" data-end=\"4816\">Deep Learning et mod\u00e8les NLP<\/h2>\n<ul data-start=\"4817\" data-end=\"5237\">\n<li data-start=\"4817\" data-end=\"5237\">\n<p data-start=\"4819\" data-end=\"5237\">Les <strong data-start=\"4823\" data-end=\"4852\">embeddings contextualis\u00e9s<\/strong> (BERT, SciBERT) ont am\u00e9lior\u00e9 la repr\u00e9sentation s\u00e9mantique des r\u00e9sum\u00e9s et titres, permettant des similarit\u00e9s plus fines entre articles. Les mod\u00e8les s\u00e9quentiels (RNN, Transformers) sont utilis\u00e9s pour mod\u00e9liser parcours\/flux d\u2019int\u00e9r\u00eat. Cependant, le co\u00fbt compute et la latence demeurent des challenges pour l\u2019int\u00e9gration temps r\u00e9el \u00e0 grande \u00e9chelle. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ul>\n<h2 data-start=\"5239\" data-end=\"5272\">Recommandation en temps r\u00e9el<\/h2>\n<ul data-start=\"5273\" data-end=\"5761\">\n<li data-start=\"5273\" data-end=\"5429\">\n<p data-start=\"5275\" data-end=\"5429\"><strong data-start=\"5275\" data-end=\"5288\">N\u00e9cessit\u00e9<\/strong> : mise \u00e0 jour instantan\u00e9e des recommandations lorsque de nouveaux articles sont index\u00e9s ou quand l\u2019utilisateur change de centre d\u2019int\u00e9r\u00eat.<\/p>\n<\/li>\n<li data-start=\"5430\" data-end=\"5761\">\n<p data-start=\"5432\" data-end=\"5761\"><strong data-start=\"5432\" data-end=\"5454\">Technologies clefs<\/strong> : Apache Kafka (ingestion\/queue), Spark Streaming \/ Flink (traitement), bases NoSQL rapides (Elasticsearch, Redis) pour indexation et requ\u00eates rapides. Ces \u00e9l\u00e9ments sont explicitement recommand\u00e9s dans les travaux r\u00e9cents portant sur plateformes acad\u00e9miques temps-r\u00e9el. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ul>\n<hr data-start=\"5763\" data-end=\"5766\" \/>\n<h1 data-start=\"5768\" data-end=\"5809\">Analyse comparative (synth\u00e8se pratique)<\/h1>\n<div class=\"TyagGW_tableContainer\">\n<div class=\"group TyagGW_tableWrapper flex w-fit flex-col-reverse\" tabindex=\"-1\">\n<table class=\"w-fit min-w-(--thread-content-width)\" data-start=\"5811\" data-end=\"6283\">\n<thead data-start=\"5811\" data-end=\"5857\">\n<tr data-start=\"5811\" data-end=\"5857\">\n<th data-start=\"5811\" data-end=\"5821\" data-col-size=\"sm\">Crit\u00e8re<\/th>\n<th data-start=\"5821\" data-end=\"5831\" data-col-size=\"sm\">Contenu<\/th>\n<th data-start=\"5831\" data-end=\"5846\" data-col-size=\"sm\">Collaboratif<\/th>\n<th data-start=\"5846\" data-end=\"5857\" data-col-size=\"sm\">Hybride<\/th>\n<\/tr>\n<\/thead>\n<tbody data-start=\"5879\" data-end=\"6283\">\n<tr data-start=\"5879\" data-end=\"5938\">\n<td data-start=\"5879\" data-end=\"5897\" data-col-size=\"sm\">Cold-start item<\/td>\n<td data-start=\"5897\" data-end=\"5918\" data-col-size=\"sm\">\u2705 (si m\u00e9tadonn\u00e9es)<\/td>\n<td data-start=\"5918\" data-end=\"5922\" data-col-size=\"sm\">\u274c<\/td>\n<td data-start=\"5922\" data-end=\"5938\" data-col-size=\"sm\">\u2705 (meilleur)<\/td>\n<\/tr>\n<tr data-start=\"5939\" data-end=\"5988\">\n<td data-start=\"5939\" data-end=\"5957\" data-col-size=\"sm\">Cold-start user<\/td>\n<td data-start=\"5957\" data-end=\"5961\" data-col-size=\"sm\">\u274c<\/td>\n<td data-start=\"5961\" data-end=\"5965\" data-col-size=\"sm\">\u274c<\/td>\n<td data-start=\"5965\" data-end=\"5988\" data-col-size=\"sm\">\u2705 (avec onboarding)<\/td>\n<\/tr>\n<tr data-start=\"5989\" data-end=\"6046\">\n<td data-start=\"5989\" data-end=\"6001\" data-col-size=\"sm\">Diversit\u00e9<\/td>\n<td data-start=\"6001\" data-end=\"6010\" data-col-size=\"sm\">Faible<\/td>\n<td data-start=\"6010\" data-end=\"6020\" data-col-size=\"sm\">Moyenne<\/td>\n<td data-start=\"6020\" data-end=\"6046\" data-col-size=\"sm\">Meilleure (si designs)<\/td>\n<\/tr>\n<tr data-start=\"6047\" data-end=\"6091\">\n<td data-start=\"6047\" data-end=\"6063\" data-col-size=\"sm\">Explicabilit\u00e9<\/td>\n<td data-start=\"6063\" data-end=\"6071\" data-col-size=\"sm\">Haute<\/td>\n<td data-start=\"6071\" data-end=\"6080\" data-col-size=\"sm\">Faible<\/td>\n<td data-start=\"6080\" data-end=\"6091\" data-col-size=\"sm\">Moyenne<\/td>\n<\/tr>\n<tr data-start=\"6092\" data-end=\"6163\">\n<td data-start=\"6092\" data-end=\"6106\" data-col-size=\"sm\">Scalabilit\u00e9<\/td>\n<td data-start=\"6106\" data-end=\"6114\" data-col-size=\"sm\">Bonne<\/td>\n<td data-start=\"6114\" data-end=\"6133\" data-col-size=\"sm\">D\u00e9pend des algos<\/td>\n<td data-start=\"6133\" data-end=\"6163\" data-col-size=\"sm\">Complexe (plus d\u2019\u00e9l\u00e9ments)<\/td>\n<\/tr>\n<tr data-start=\"6164\" data-end=\"6283\">\n<td data-start=\"6164\" data-end=\"6185\" data-col-size=\"sm\">Latence temps r\u00e9el<\/td>\n<td data-start=\"6185\" data-end=\"6210\" data-col-size=\"sm\">Bonne (index + search)<\/td>\n<td data-start=\"6210\" data-end=\"6240\" data-col-size=\"sm\">Peut \u00eatre co\u00fbteux (updates)<\/td>\n<td data-start=\"6240\" data-end=\"6283\" data-col-size=\"sm\">D\u00e9fi : orchestrer pipelines temps r\u00e9el.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<p data-start=\"6285\" data-end=\"6814\">Explications : pour les plateformes acad\u00e9miques o\u00f9 les m\u00e9tadonn\u00e9es (titre, r\u00e9sum\u00e9, auteurs, affili-) sont disponibles, le contenu assure un socle robuste. Le collaboratif devient puissant apr\u00e8s constitution d\u2019un large historique d\u2019utilisateurs. Les meilleurs r\u00e9sultats naissent des architectures hybrides qui pond\u00e8rent dynamiquement les scores selon la situation (nouvel utilisateur \u2192 privil\u00e9gier contenu; utilisateur \u00e9tabli \u2192 pousser collaboratif). <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<p class=\"not-prose mt-0! mb-0! flex-auto truncate\">ETAT DE L&#8217;ART SUR LE SUJET DU\u2026<\/p>\n<p class=\"not-prose mt-0! mb-0! flex-auto truncate\">ETAT DE L&#8217;ART SUR LE SUJET DU\u2026<\/p>\n<hr data-start=\"6816\" data-end=\"6819\" \/>\n<h1 data-start=\"6821\" data-end=\"6887\">Enjeux techniques et architecturaux pour plateformes acad\u00e9miques<\/h1>\n<h2 data-start=\"6889\" data-end=\"6932\">Donn\u00e9es \u00e0 collecter (minimum recommand\u00e9)<\/h2>\n<ul>\n<li data-start=\"6935\" data-end=\"6998\">M\u00e9tadonn\u00e9es: titre, auteurs, r\u00e9sum\u00e9, mots-cl\u00e9s, date, source.<\/li>\n<li data-start=\"7001\" data-end=\"7053\">Texte complet si accessible (am\u00e9liore embeddings).<\/li>\n<li data-start=\"7056\" data-end=\"7144\">Interactions: clics, dur\u00e9e de lecture, t\u00e9l\u00e9chargements, partages, favoris, recherches.<\/li>\n<li data-start=\"7147\" data-end=\"7263\">Signals additionnels: affiliations, discipline, niveau (\u00e9tudiant\/chercheur).<button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/li>\n<\/ul>\n<h2 data-start=\"7265\" data-end=\"7302\">Pipeline recommand\u00e9 (haute-niveau)<\/h2>\n<ol data-start=\"7303\" data-end=\"8146\">\n<li data-start=\"7303\" data-end=\"7390\">\n<p data-start=\"7306\" data-end=\"7390\"><strong data-start=\"7306\" data-end=\"7319\">Ingestion<\/strong> : flux d\u2019articles via API (arXiv, PubMed, Semantic Scholar) \u2192 Kafka.<\/p>\n<\/li>\n<li data-start=\"7391\" data-end=\"7516\">\n<p data-start=\"7394\" data-end=\"7516\"><strong data-start=\"7394\" data-end=\"7411\">Pr\u00e9traitement<\/strong> : extraction m\u00e9tadonn\u00e9es + nettoyage + cr\u00e9ation d\u2019embeddings (BERT\/SciBERT) en mode batch incr\u00e9mental.<\/p>\n<\/li>\n<li data-start=\"7517\" data-end=\"7629\">\n<p data-start=\"7520\" data-end=\"7629\"><strong data-start=\"7520\" data-end=\"7532\">Stockage<\/strong> : indexation dans Elasticsearch (recherche par texte) + base NoSQL (Redis) pour scores chauds.<\/p>\n<\/li>\n<li data-start=\"7630\" data-end=\"7880\">\n<p data-start=\"7633\" data-end=\"7652\"><strong data-start=\"7633\" data-end=\"7649\">Mod\u00e9lisation<\/strong> :<\/p>\n<\/li>\n<\/ol>\n<ul>\n<li style=\"list-style-type: none;\">\n<ul>\n<li data-start=\"7658\" data-end=\"7706\">Content-based: similarit\u00e9 embeddings + TF-IDF.<\/li>\n<li data-start=\"7712\" data-end=\"7796\">Collaborative: factorisation matricielle \/ embeddings implicites (ALS, neural CF).<\/li>\n<li data-start=\"7802\" data-end=\"7880\">Orchestrateur hybride : r\u00e8gle \/ mod\u00e8le d\u2019agr\u00e9gation (pond\u00e9ration dynamique).<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<ol data-start=\"7303\" data-end=\"8146\">\n<li data-start=\"7881\" data-end=\"7997\">\n<p data-start=\"7884\" data-end=\"7997\"><strong data-start=\"7884\" data-end=\"7895\">Serving<\/strong> : endpoint de recommandations (latence &lt; 200\u2013500 ms cible), mise \u00e0 jour incr\u00e9mentale via streaming.<\/p>\n<\/li>\n<li data-start=\"7998\" data-end=\"8146\">\n<p data-start=\"8001\" data-end=\"8146\"><strong data-start=\"8001\" data-end=\"8031\">Monitoring &amp; feedback loop<\/strong> : collecte m\u00e9triques (CTR, temps sur article), r\u00e9entra\u00eener p\u00e9riodiquement.<button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ol>\n<h2 data-start=\"8148\" data-end=\"8171\">Mesures d\u2019\u00e9valuation<\/h2>\n<ul data-start=\"8172\" data-end=\"8537\">\n<li data-start=\"8172\" data-end=\"8226\">\n<p data-start=\"8174\" data-end=\"8226\"><strong data-start=\"8174\" data-end=\"8188\">Pertinence<\/strong> : pr\u00e9cision@k, rappel@k, MAP, NDCG.<\/p>\n<\/li>\n<li data-start=\"8227\" data-end=\"8292\">\n<p data-start=\"8229\" data-end=\"8292\"><strong data-start=\"8229\" data-end=\"8246\">Utilisabilit\u00e9<\/strong> : taux de clic, dur\u00e9e de lecture, partages.<\/p>\n<\/li>\n<li data-start=\"8293\" data-end=\"8537\">\n<p data-start=\"8295\" data-end=\"8537\"><strong data-start=\"8295\" data-end=\"8311\">Op\u00e9rationnel<\/strong> : latence de r\u00e9ponse, co\u00fbt de calcul, taux d\u2019erreur.<br data-start=\"8364\" data-end=\"8367\" \/>Les travaux existants \u00e9valuent fr\u00e9quemment pr\u00e9cision\/rappel et insistent sur l\u2019importance des tests sur jeux r\u00e9els d\u2019interactions. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ul>\n<hr data-start=\"8539\" data-end=\"8542\" \/>\n<h1 data-start=\"8544\" data-end=\"8583\">Recommandations pratiques et \u00e9thiques<\/h1>\n<ol data-start=\"8585\" data-end=\"9473\">\n<li data-start=\"8585\" data-end=\"8744\">\n<p data-start=\"8588\" data-end=\"8744\"><strong data-start=\"8588\" data-end=\"8623\">Commencer par un hybride simple<\/strong> : contenu + pond\u00e9ration de popularit\u00e9, \u00e9voluer vers mod\u00e8les plus sophistiqu\u00e9s. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"8745\" data-end=\"8904\">\n<p data-start=\"8748\" data-end=\"8904\"><strong data-start=\"8748\" data-end=\"8785\">Utiliser embeddings scientifiques<\/strong> (SciBERT) pour de meilleures similarit\u00e9s s\u00e9mantiques sur textes acad\u00e9miques. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"8905\" data-end=\"9095\">\n<p data-start=\"8908\" data-end=\"9095\"><strong data-start=\"8908\" data-end=\"8945\">Impl\u00e9menter un pipeline streaming<\/strong> (Kafka + Spark\/Flink) pour ing\u00e9rer et indexer nouveaux articles en continu si l\u2019objectif est le temps r\u00e9el. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"9096\" data-end=\"9309\">\n<p data-start=\"9099\" data-end=\"9309\"><strong data-start=\"9099\" data-end=\"9125\">Prot\u00e9ger la vie priv\u00e9e<\/strong> : anonymisation des logs, options d\u2019opt-out, stockage s\u00e9curis\u00e9. (Limites du collaboratif sur confidentialit\u00e9 soulign\u00e9es dans la litt\u00e9rature.) <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"9310\" data-end=\"9473\">\n<p data-start=\"9313\" data-end=\"9473\"><strong data-start=\"9313\" data-end=\"9341\">Favoriser l\u2019acc\u00e8s ouvert<\/strong> : prioriser sources OA (arXiv, PubMed Central) pour maximiser l\u2019utilit\u00e9 pour la communaut\u00e9. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ol>\n<hr data-start=\"9475\" data-end=\"9478\" \/>\n<h1 data-start=\"9480\" data-end=\"9534\">R\u00e9f\u00e9rences (s\u00e9lectives, extraites du corpus analys\u00e9)<\/h1>\n<blockquote data-start=\"9536\" data-end=\"9615\">\n<p data-start=\"9538\" data-end=\"9615\">R\u00e9f\u00e9rences cit\u00e9es dans la revue et disponibles dans l\u2019\u00e9tat de l\u2019art analys\u00e9 :<\/p>\n<\/blockquote>\n<ul data-start=\"9616\" data-end=\"10569\">\n<li data-start=\"9616\" data-end=\"9771\">\n<p data-start=\"9618\" data-end=\"9771\">M. Benkhouya &amp; Ait Abdelmalek R. (2020). <em data-start=\"9659\" data-end=\"9728\">Syst\u00e8mes de recommandation personnalis\u00e9 en recherche d\u2019informations<\/em>. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"9772\" data-end=\"9932\">\n<p data-start=\"9774\" data-end=\"9932\">DJEBARNIA, N. E. I. (2022). <em data-start=\"9802\" data-end=\"9889\">Syst\u00e8mes de recommandation des ressources en se basant sur les profils des apprenants<\/em>. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"9933\" data-end=\"10106\">\n<p data-start=\"9935\" data-end=\"10106\">Pinos Ullauri, L. A., &amp; Lebis, A. (2023). <em data-start=\"9977\" data-end=\"10063\">Syst\u00e8me de recommandation de cours bas\u00e9 sur les soft skills : algorithmes g\u00e9n\u00e9tiques<\/em>. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"10107\" data-end=\"10257\">\n<p data-start=\"10109\" data-end=\"10257\">Fatima ZOhra, Bouroumi, Atika (2021). <em data-start=\"10147\" data-end=\"10214\">Syst\u00e8me de recommandation bas\u00e9s sur Deep Learning dans le E-Sant\u00e9<\/em>. <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"10258\" data-end=\"10426\">\n<p data-start=\"10260\" data-end=\"10426\">Mercanti-Gu\u00e9rin, M. (2014). <em data-start=\"10288\" data-end=\"10384\">Syst\u00e8mes de recommandation et r\u00e9seaux sociaux, quelles implications pour le marketing digital?<\/em> <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<li data-start=\"10427\" data-end=\"10569\">\n<p data-start=\"10429\" data-end=\"10569\">ArZiGo: <em data-start=\"10437\" data-end=\"10486\">A Recommendation System for Scientific Articles<\/em> (prototype cit\u00e9 dans l\u2019\u00e9tude comparative). <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<\/li>\n<\/ul>\n<p data-start=\"10571\" data-end=\"10892\"><em data-start=\"10571\" data-end=\"10852\">(Les extraits, analyses et recommandations ci-dessus s\u2019appuient principalement sur le m\u00e9moire et l\u2019\u00e9tat de l\u2019art que vous avez fourni, qui synth\u00e9tise ces travaux et propose une architecture prototype pour une plateforme de recommandation d\u2019articles scientifiques en temps r\u00e9el).)<\/em> <button class=\"ms-1 flex h-[25px] text-[10px] leading-[13px] rounded-xl corner-superellipse\/1.1 items-center justify-center gap-1 px-2 relative text-token-text-secondary! hover:text-token-text-primary! hover:bg-token-bg-secondary dark:bg-token-main-surface-secondary dark:hover:bg-token-bg-secondary bg-[#f4f4f4] \"><\/button><\/p>\n<hr data-start=\"10894\" data-end=\"10897\" \/>\n<h1 data-start=\"10899\" data-end=\"10913\">Conclusion<\/h1>\n<p data-start=\"10914\" data-end=\"11558\">Les syst\u00e8mes de recommandation IA offrent un levier puissant pour am\u00e9liorer la d\u00e9couverte scientifique. Pour les plateformes acad\u00e9miques, l\u2019approche hybride \u2014 renforc\u00e9e par des embeddings scientifiques et un pipeline temps r\u00e9el \u2014 appara\u00eet comme le meilleur compromis entre pertinence, r\u00e9activit\u00e9 et couverture. Toutefois, la conception doit \u00e9quilibrer performances techniques, co\u00fbt, explicabilit\u00e9 et respect de la vie priv\u00e9e. Le m\u00e9moire analys\u00e9 constitue une feuille de route op\u00e9rationnelle pour impl\u00e9menter un prototype temps r\u00e9el et fournit des points d\u2019appui m\u00e9thodologiques et technologiques solides.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>R\u00e9sum\u00e9 (fran\u00e7ais) La croissance explosive de la production scientifique rend la d\u00e9couverte d\u2019articles pertinents de plus en plus difficile. Les syst\u00e8mes de recommandation \u2014 d\u00e9j\u00e0 efficaces dans le commerce et les m\u00e9dias \u2014 constituent une solution prometteuse pour les plateformes acad\u00e9miques. Cet article propose une synth\u00e8se compl\u00e8te : r\u00e9sum\u00e9, abstract, introduction, \u00e9tat de l\u2019art (revue [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":2048,"menu_order":3,"comment_status":"closed","ping_status":"closed","template":"","doc_tag":[],"class_list":["post-2101","docs","type-docs","status-publish","hentry","no-post-thumbnail"],"acf":[],"_links":{"self":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/docs\/2101","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/docs"}],"about":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/types\/docs"}],"author":[{"embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/comments?post=2101"}],"version-history":[{"count":2,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/docs\/2101\/revisions"}],"predecessor-version":[{"id":6596,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/docs\/2101\/revisions\/6596"}],"up":[{"embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/docs\/2048"}],"wp:attachment":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/media?parent=2101"}],"wp:term":[{"taxonomy":"doc_tag","embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/doc_tag?post=2101"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}