{"id":6248,"date":"2025-12-11T10:44:33","date_gmt":"2025-12-11T10:44:33","guid":{"rendered":"https:\/\/sahelib.atatec-design.com\/index.php\/2025\/12\/11\/modeles-hybrides-nlp-graphes-pour-la-decouverte-scientifique\/"},"modified":"2025-12-11T11:14:49","modified_gmt":"2025-12-11T11:14:49","slug":"modeles-hybrides-nlp-graphes-pour-la-decouverte-scientifique","status":"publish","type":"post","link":"https:\/\/sahelib.atatec-design.com\/index.php\/2025\/12\/11\/modeles-hybrides-nlp-graphes-pour-la-decouverte-scientifique\/","title":{"rendered":"Mod\u00e8les hybrides NLP-graphes pour la d\u00e9couverte scientifique"},"content":{"rendered":"<h2>Mod\u00e8les hybrides NLP-graphes pour la d\u00e9couverte scientifique<\/h2>\n<p><strong>Auteur(s) :<\/strong> Dr. Jean Moreau \u2014 <strong>Date :<\/strong> 2023-07-02 \u2014 <strong>Source :<\/strong> Semantic Scholar<\/p>\n<h1 data-start=\"685\" data-end=\"708\"><strong data-start=\"687\" data-end=\"708\">R\u00e9sum\u00e9 (Abstract)<\/strong><\/h1>\n<p data-start=\"710\" data-end=\"1638\">L\u2019explosion de la production scientifique rend la d\u00e9couverte de connaissances pertinentes de plus en plus complexe. Les mod\u00e8les hybrides combinant le traitement du langage naturel (NLP) et les graphes de connaissances offrent une solution innovante pour explorer, relier et interpr\u00e9ter l\u2019information scientifique \u00e0 grande \u00e9chelle. Cet article propose une revue d\u00e9taill\u00e9e des approches hybrides NLP-graphes, pr\u00e9sentant les techniques d\u2019extraction d\u2019entit\u00e9s, d\u2019analyse s\u00e9mantique, de repr\u00e9sentation vectorielle et de structuration des connaissances sous forme de graphes. Nous comparons les m\u00e9thodes existantes, discutons de leurs applications dans la recommandation d\u2019articles, la d\u00e9tection de tendances scientifiques et la d\u00e9couverte de corr\u00e9lations interdisciplinaires. Enfin, nous mettons en \u00e9vidence les d\u00e9fis actuels, les limites des mod\u00e8les et les perspectives pour l\u2019optimisation de la d\u00e9couverte scientifique automatis\u00e9e.<\/p>\n<p data-start=\"1640\" data-end=\"1792\"><strong data-start=\"1640\" data-end=\"1655\">Mots-cl\u00e9s :<\/strong> NLP, graphes de connaissances, d\u00e9couverte scientifique, apprentissage automatique, intelligence artificielle, extraction d\u2019informations.<\/p>\n<p data-start=\"1794\" data-end=\"1818\"><strong data-start=\"1794\" data-end=\"1816\">Abstract (English)<\/strong><\/p>\n<p data-start=\"1820\" data-end=\"2512\">The rapid growth of scientific publications increases the complexity of identifying relevant knowledge. Hybrid models combining Natural Language Processing (NLP) and knowledge graphs provide an innovative approach to explore, link, and interpret large-scale scientific data. This paper presents a comprehensive review of NLP-graph hybrid methods, covering entity extraction, semantic analysis, vector representation, and graph structuring. We compare existing approaches and discuss applications in article recommendation, trend detection, and interdisciplinary knowledge discovery. Challenges, limitations, and future directions for optimizing automated scientific discovery are highlighted.<\/p>\n<p data-start=\"2514\" data-end=\"2639\"><strong data-start=\"2514\" data-end=\"2527\">Keywords:<\/strong> NLP, knowledge graphs, scientific discovery, machine learning, artificial intelligence, information extraction.<\/p>\n<hr data-start=\"2641\" data-end=\"2644\" \/>\n<h1 data-start=\"2646\" data-end=\"2664\"><strong data-start=\"2648\" data-end=\"2664\">Introduction<\/strong><\/h1>\n<p data-start=\"2666\" data-end=\"3059\">La recherche scientifique contemporaine est caract\u00e9ris\u00e9e par une explosion de publications dans tous les domaines. Cette prolif\u00e9ration rend le tri et l\u2019analyse de l\u2019information manuelle difficile et chronophage. Dans ce contexte, les <strong data-start=\"2900\" data-end=\"2945\">mod\u00e8les hybrides combinant NLP et graphes<\/strong> apparaissent comme une solution prometteuse pour organiser et exploiter efficacement les donn\u00e9es scientifiques.<\/p>\n<ul data-start=\"3061\" data-end=\"3390\">\n<li data-start=\"3061\" data-end=\"3218\">\n<p data-start=\"3063\" data-end=\"3218\"><strong data-start=\"3063\" data-end=\"3100\">NLP (Natural Language Processing)<\/strong> permet l\u2019extraction d\u2019informations cl\u00e9s \u00e0 partir de textes, comme les entit\u00e9s, relations et concepts scientifiques.<\/p>\n<\/li>\n<li data-start=\"3219\" data-end=\"3390\">\n<p data-start=\"3221\" data-end=\"3390\"><strong data-start=\"3221\" data-end=\"3249\">Graphes de connaissances<\/strong> repr\u00e9sentent ces informations sous forme de r\u00e9seaux structur\u00e9s, facilitant les liens entre concepts, auteurs, publications et disciplines.<\/p>\n<\/li>\n<\/ul>\n<p data-start=\"3392\" data-end=\"3628\">L\u2019approche hybride consiste \u00e0 combiner la <strong data-start=\"3434\" data-end=\"3464\">capacit\u00e9 s\u00e9mantique du NLP<\/strong> avec la <strong data-start=\"3473\" data-end=\"3516\">structuration relationnelle des graphes<\/strong>, permettant ainsi d\u2019identifier de nouvelles corr\u00e9lations, tendances et opportunit\u00e9s de d\u00e9couverte scientifique.<\/p>\n<p data-start=\"3630\" data-end=\"3885\">L\u2019objectif de cet article est de fournir une <strong data-start=\"3675\" data-end=\"3730\">revue syst\u00e9matique des mod\u00e8les hybrides NLP-graphes<\/strong>, de pr\u00e9senter les m\u00e9thodes et applications r\u00e9centes, d\u2019analyser comparativement leur performance et de proposer des perspectives pour la recherche future.<\/p>\n<hr data-start=\"3887\" data-end=\"3890\" \/>\n<h1 data-start=\"3892\" data-end=\"3932\"><strong data-start=\"3894\" data-end=\"3932\">\u00c9tat de l\u2019art \/ Revue syst\u00e9matique<\/strong><\/h1>\n<ol data-start=\"3934\" data-end=\"5321\">\n<li data-start=\"3934\" data-end=\"4415\">\n<p data-start=\"3937\" data-end=\"3984\"><strong data-start=\"3937\" data-end=\"3984\">Mod\u00e8les NLP pour la d\u00e9couverte scientifique<\/strong><\/p>\n<ul data-start=\"3988\" data-end=\"4415\">\n<li data-start=\"3988\" data-end=\"4090\">\n<p data-start=\"3990\" data-end=\"4090\">Extraction d\u2019entit\u00e9s nomm\u00e9es (NER) : identification des termes scientifiques, auteurs, institutions.<\/p>\n<\/li>\n<li data-start=\"4094\" data-end=\"4178\">\n<p data-start=\"4096\" data-end=\"4178\">Analyse de relations s\u00e9mantiques : co-occurrences, liens causaux ou collaboratifs.<\/p>\n<\/li>\n<li data-start=\"4182\" data-end=\"4304\">\n<p data-start=\"4184\" data-end=\"4304\">Embeddings et repr\u00e9sentations vectorielles : Word2Vec, BERT, SciBERT, BioBERT pour capturer les relations contextuelles.<\/p>\n<\/li>\n<li data-start=\"4308\" data-end=\"4415\">\n<p data-start=\"4310\" data-end=\"4415\">Limites : manque de structure relationnelle explicite, difficult\u00e9 \u00e0 relier des concepts \u00e0 grande \u00e9chelle.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"4417\" data-end=\"4829\">\n<p data-start=\"4420\" data-end=\"4448\"><strong data-start=\"4420\" data-end=\"4448\">Graphes de connaissances<\/strong><\/p>\n<ul data-start=\"4452\" data-end=\"4829\">\n<li data-start=\"4452\" data-end=\"4536\">\n<p data-start=\"4454\" data-end=\"4536\">D\u00e9finition : structures o\u00f9 les n\u0153uds repr\u00e9sentent entit\u00e9s et les ar\u00eates relations.<\/p>\n<\/li>\n<li data-start=\"4540\" data-end=\"4630\">\n<p data-start=\"4542\" data-end=\"4630\">Applications : Knowledge Graphs scientifiques (ex. Microsoft Academic Graph, PubMed KG).<\/p>\n<\/li>\n<li data-start=\"4634\" data-end=\"4735\">\n<p data-start=\"4636\" data-end=\"4735\">Avantages : repr\u00e9sentation explicite des relations, support des inf\u00e9rences, d\u00e9couverte de patterns.<\/p>\n<\/li>\n<li data-start=\"4739\" data-end=\"4829\">\n<p data-start=\"4741\" data-end=\"4829\">Limites : construction co\u00fbteuse, sparsit\u00e9 des graphes, int\u00e9gration des textes complexes.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"4831\" data-end=\"5321\">\n<p data-start=\"4834\" data-end=\"4868\"><strong data-start=\"4834\" data-end=\"4868\">Approches hybrides NLP-graphes<\/strong><\/p>\n<ul data-start=\"4872\" data-end=\"5321\">\n<li data-start=\"4872\" data-end=\"4949\">\n<p data-start=\"4874\" data-end=\"4949\">Extraction automatique de graphes \u00e0 partir de textes scientifiques via NLP.<\/p>\n<\/li>\n<li data-start=\"4953\" data-end=\"5034\">\n<p data-start=\"4955\" data-end=\"5034\">Repr\u00e9sentation de concepts scientifiques et de relations sous forme de graphes.<\/p>\n<\/li>\n<li data-start=\"5038\" data-end=\"5177\">\n<p data-start=\"5040\" data-end=\"5177\">Algorithmes de propagation et embedding de graphes (Graph Neural Networks, GraphSAGE, GAT) pour enrichir les repr\u00e9sentations s\u00e9mantiques.<\/p>\n<\/li>\n<li data-start=\"5181\" data-end=\"5321\">\n<p data-start=\"5183\" data-end=\"5321\">R\u00e9sultats : meilleure d\u00e9tection des liens implicites, am\u00e9lioration de la recommandation d\u2019articles et de la d\u00e9couverte interdisciplinaire.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<hr data-start=\"5323\" data-end=\"5326\" \/>\n<h1 data-start=\"5328\" data-end=\"5366\"><strong data-start=\"5330\" data-end=\"5366\">M\u00e9thodologie et mod\u00e8les hybrides<\/strong><\/h1>\n<ol data-start=\"5368\" data-end=\"6400\">\n<li data-start=\"5368\" data-end=\"5828\">\n<p data-start=\"5371\" data-end=\"5391\"><strong data-start=\"5371\" data-end=\"5391\">Pipeline typique<\/strong><\/p>\n<ul data-start=\"5395\" data-end=\"5828\">\n<li data-start=\"5395\" data-end=\"5468\">\n<p data-start=\"5397\" data-end=\"5468\">Pr\u00e9traitement : tokenisation, normalisation, suppression des stopwords.<\/p>\n<\/li>\n<li data-start=\"5472\" data-end=\"5516\">\n<p data-start=\"5474\" data-end=\"5516\">Extraction d\u2019entit\u00e9s et relations via NLP.<\/p>\n<\/li>\n<li data-start=\"5520\" data-end=\"5631\">\n<p data-start=\"5522\" data-end=\"5631\">Construction du graphe : n\u0153uds (entit\u00e9s), ar\u00eates (relations), pond\u00e9rations bas\u00e9es sur fr\u00e9quence ou confiance.<\/p>\n<\/li>\n<li data-start=\"5635\" data-end=\"5723\">\n<p data-start=\"5637\" data-end=\"5723\">Apprentissage sur graphes : embeddings, algorithmes de recherche de motifs, inf\u00e9rence.<\/p>\n<\/li>\n<li data-start=\"5727\" data-end=\"5828\">\n<p data-start=\"5729\" data-end=\"5828\">Recommandation ou d\u00e9couverte : identification de relations inattendues ou de nouvelles th\u00e9matiques.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"5830\" data-end=\"6159\">\n<p data-start=\"5833\" data-end=\"5859\"><strong data-start=\"5833\" data-end=\"5859\">Techniques principales<\/strong><\/p>\n<ul data-start=\"5863\" data-end=\"6159\">\n<li data-start=\"5863\" data-end=\"5952\">\n<p data-start=\"5865\" data-end=\"5952\"><strong data-start=\"5865\" data-end=\"5891\">NLP profond (Deep NLP)<\/strong> : BERT, SciBERT, BioBERT pour contextualisation des entit\u00e9s.<\/p>\n<\/li>\n<li data-start=\"5956\" data-end=\"6040\">\n<p data-start=\"5958\" data-end=\"6040\"><strong data-start=\"5958\" data-end=\"5978\">Graph Embeddings<\/strong> : Node2Vec, GraphSAGE, GAT pour propagation des informations.<\/p>\n<\/li>\n<li data-start=\"6044\" data-end=\"6159\">\n<p data-start=\"6046\" data-end=\"6159\"><strong data-start=\"6046\" data-end=\"6061\">Hybridation<\/strong> : combinaison des embeddings NLP et des embeddings graphes pour obtenir des vecteurs plus riches.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"6161\" data-end=\"6400\">\n<p data-start=\"6164\" data-end=\"6199\"><strong data-start=\"6164\" data-end=\"6199\">Exemples d\u2019outils et frameworks<\/strong><\/p>\n<ul data-start=\"6203\" data-end=\"6400\">\n<li data-start=\"6203\" data-end=\"6256\">\n<p data-start=\"6205\" data-end=\"6256\">SpaCy, AllenNLP, HuggingFace Transformers pour NLP.<\/p>\n<\/li>\n<li data-start=\"6260\" data-end=\"6326\">\n<p data-start=\"6262\" data-end=\"6326\">Neo4j, DGL (Deep Graph Library), PyTorch Geometric pour graphes.<\/p>\n<\/li>\n<li data-start=\"6330\" data-end=\"6400\">\n<p data-start=\"6332\" data-end=\"6400\">Plateformes hybrides : Microsoft Academic Knowledge Graph, OpenAlex.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<hr data-start=\"6402\" data-end=\"6405\" \/>\n<h1 data-start=\"6407\" data-end=\"6440\"><strong data-start=\"6409\" data-end=\"6440\">Applications et cas d\u2019usage<\/strong><\/h1>\n<ol data-start=\"6442\" data-end=\"6939\">\n<li data-start=\"6442\" data-end=\"6615\">\n<p data-start=\"6445\" data-end=\"6476\"><strong data-start=\"6445\" data-end=\"6476\">Recommandation scientifique<\/strong><\/p>\n<ul data-start=\"6480\" data-end=\"6615\">\n<li data-start=\"6480\" data-end=\"6555\">\n<p data-start=\"6482\" data-end=\"6555\">Suggestion d\u2019articles pertinents selon profils et historiques de lecture.<\/p>\n<\/li>\n<li data-start=\"6559\" data-end=\"6615\">\n<p data-start=\"6561\" data-end=\"6615\">D\u00e9tection de concepts connexes et interdisciplinaires.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"6617\" data-end=\"6784\">\n<p data-start=\"6620\" data-end=\"6644\"><strong data-start=\"6620\" data-end=\"6644\">Analyse de tendances<\/strong><\/p>\n<ul data-start=\"6648\" data-end=\"6784\">\n<li data-start=\"6648\" data-end=\"6729\">\n<p data-start=\"6650\" data-end=\"6729\">Suivi des \u00e9mergences th\u00e9matiques, identification des sujets \u00e0 forte croissance.<\/p>\n<\/li>\n<li data-start=\"6733\" data-end=\"6784\">\n<p data-start=\"6735\" data-end=\"6784\">Pr\u00e9vision des domaines scientifiques prometteurs.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"6786\" data-end=\"6939\">\n<p data-start=\"6789\" data-end=\"6830\"><strong data-start=\"6789\" data-end=\"6830\">D\u00e9couverte de corr\u00e9lations implicites<\/strong><\/p>\n<ul data-start=\"6834\" data-end=\"6939\">\n<li data-start=\"6834\" data-end=\"6939\">\n<p data-start=\"6836\" data-end=\"6939\">Identification de relations entre concepts ou auteurs non directement apparentes dans les publications.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<hr data-start=\"6941\" data-end=\"6944\" \/>\n<h1 data-start=\"6946\" data-end=\"6971\"><strong data-start=\"6948\" data-end=\"6971\">Analyse comparative<\/strong><\/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=\"6973\" data-end=\"7436\">\n<thead data-start=\"6973\" data-end=\"7007\">\n<tr data-start=\"6973\" data-end=\"7007\">\n<th data-start=\"6973\" data-end=\"6984\" data-col-size=\"sm\">Approche<\/th>\n<th data-start=\"6984\" data-end=\"6996\" data-col-size=\"md\">Avantages<\/th>\n<th data-start=\"6996\" data-end=\"7007\" data-col-size=\"md\">Limites<\/th>\n<\/tr>\n<\/thead>\n<tbody data-start=\"7043\" data-end=\"7436\">\n<tr data-start=\"7043\" data-end=\"7153\">\n<td data-start=\"7043\" data-end=\"7053\" data-col-size=\"sm\">NLP pur<\/td>\n<td data-start=\"7053\" data-end=\"7104\" data-col-size=\"md\">Extraction rapide des entit\u00e9s, contextualisation<\/td>\n<td data-start=\"7104\" data-end=\"7153\" data-col-size=\"md\">Pas de repr\u00e9sentation explicite des relations<\/td>\n<\/tr>\n<tr data-start=\"7154\" data-end=\"7238\">\n<td data-start=\"7154\" data-end=\"7169\" data-col-size=\"sm\">Graphes purs<\/td>\n<td data-start=\"7169\" data-end=\"7202\" data-col-size=\"md\">Relation explicite, inf\u00e9rences<\/td>\n<td data-start=\"7202\" data-end=\"7238\" data-col-size=\"md\">Construction lourde, maintenance<\/td>\n<\/tr>\n<tr data-start=\"7239\" data-end=\"7436\">\n<td data-start=\"7239\" data-end=\"7261\" data-col-size=\"sm\">Hybride NLP-graphes<\/td>\n<td data-start=\"7261\" data-end=\"7348\" data-col-size=\"md\">Capture du contexte et des relations, meilleure recommandation, d\u00e9couverte implicite<\/td>\n<td data-start=\"7348\" data-end=\"7436\" data-col-size=\"md\">Complexit\u00e9 computationnelle, besoin de donn\u00e9es massives et de standards pour graphes<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<p data-start=\"7438\" data-end=\"7573\">L\u2019approche hybride surpasse les m\u00e9thodes individuelles en termes de <strong data-start=\"7506\" data-end=\"7572\">pr\u00e9cision, rappel et capacit\u00e9 \u00e0 d\u00e9couvrir des liens implicites<\/strong>.<\/p>\n<hr data-start=\"7575\" data-end=\"7578\" \/>\n<h1 data-start=\"7580\" data-end=\"7607\"><strong data-start=\"7582\" data-end=\"7607\">Discussion et limites<\/strong><\/h1>\n<ul data-start=\"7609\" data-end=\"7973\">\n<li data-start=\"7609\" data-end=\"7787\">\n<p data-start=\"7611\" data-end=\"7618\">D\u00e9fis :<\/p>\n<ul data-start=\"7621\" data-end=\"7787\">\n<li data-start=\"7621\" data-end=\"7662\">\n<p data-start=\"7623\" data-end=\"7662\">Complexit\u00e9 computationnelle et m\u00e9moire.<\/p>\n<\/li>\n<li data-start=\"7665\" data-end=\"7724\">\n<p data-start=\"7667\" data-end=\"7724\">Int\u00e9gration de sources h\u00e9t\u00e9rog\u00e8nes (PubMed, ArXiv, IEEE).<\/p>\n<\/li>\n<li data-start=\"7727\" data-end=\"7787\">\n<p data-start=\"7729\" data-end=\"7787\">Evaluation difficile : absence de gold standard universel.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li data-start=\"7788\" data-end=\"7973\">\n<p data-start=\"7790\" data-end=\"7804\">Perspectives :<\/p>\n<ul data-start=\"7807\" data-end=\"7973\">\n<li data-start=\"7807\" data-end=\"7846\">\n<p data-start=\"7809\" data-end=\"7846\">Optimisation des embeddings hybrides.<\/p>\n<\/li>\n<li data-start=\"7849\" data-end=\"7893\">\n<p data-start=\"7851\" data-end=\"7893\">Standardisation des graphes scientifiques.<\/p>\n<\/li>\n<li data-start=\"7896\" data-end=\"7973\">\n<p data-start=\"7898\" data-end=\"7973\">Applications \u00e0 la veille strat\u00e9gique et \u00e0 la d\u00e9couverte interdisciplinaire.<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr data-start=\"7975\" data-end=\"7978\" \/>\n<h1 data-start=\"7980\" data-end=\"8012\"><strong data-start=\"7982\" data-end=\"8012\">Conclusion et perspectives<\/strong><\/h1>\n<p data-start=\"8014\" data-end=\"8522\">Les mod\u00e8les hybrides NLP-graphes repr\u00e9sentent une avanc\u00e9e majeure pour la d\u00e9couverte scientifique automatis\u00e9e. Ils combinent la richesse s\u00e9mantique des textes avec la structuration relationnelle des graphes, permettant de r\u00e9v\u00e9ler des liens implicites et de soutenir la recommandation d\u2019articles, la d\u00e9tection de tendances et l\u2019exploration interdisciplinaire. Les travaux futurs devraient se concentrer sur l\u2019optimisation des mod\u00e8les, l\u2019int\u00e9gration multi-sources et l\u2019\u00e9valuation standardis\u00e9e des performances.<\/p>\n<hr data-start=\"8524\" data-end=\"8527\" \/>\n<h1 data-start=\"8529\" data-end=\"8570\"><strong data-start=\"8531\" data-end=\"8570\">R\u00e9f\u00e9rences (s\u00e9lection scientifique)<\/strong><\/h1>\n<ol data-start=\"8572\" data-end=\"9356\">\n<li data-start=\"8572\" data-end=\"8765\">\n<p data-start=\"8575\" data-end=\"8765\">Wang, Q., Mao, Z., Wang, B., &amp; Guo, L. (2017). Knowledge Graph Embedding: A Survey of Approaches and Applications. <em data-start=\"8690\" data-end=\"8747\">IEEE Transactions on Knowledge and Data Engineering, 29<\/em>(12), 2724\u20132743.<\/p>\n<\/li>\n<li data-start=\"8766\" data-end=\"8878\">\n<p data-start=\"8769\" data-end=\"8878\">Beltagy, I., Lo, K., &amp; Cohan, A. (2019). SciBERT: A Pretrained Language Model for Scientific Text. <em data-start=\"8868\" data-end=\"8875\">EMNLP<\/em>.<\/p>\n<\/li>\n<li data-start=\"8879\" data-end=\"9042\">\n<p data-start=\"8882\" data-end=\"9042\">Hamilton, W. L., Ying, Z., &amp; Leskovec, J. (2017). Representation Learning on Graphs: Methods and Applications. <em data-start=\"8993\" data-end=\"9029\">IEEE Data Engineering Bulletin, 40<\/em>(3), 52\u201374.<\/p>\n<\/li>\n<li data-start=\"9043\" data-end=\"9206\">\n<p data-start=\"9046\" data-end=\"9206\">Zhou, J., Cui, G., Zhang, Z., Yang, C., Liu, Z., Wang, L., \u2026 Sun, M. (2020). Graph Neural Networks: A Review of Methods and Applications. <em data-start=\"9184\" data-end=\"9196\">AI Open, 1<\/em>, 57\u201381.<\/p>\n<\/li>\n<li data-start=\"9207\" data-end=\"9356\">\n<p data-start=\"9210\" data-end=\"9356\">Shin, H. C., &amp; Radev, D. (2020). Knowledge Graph Construction for Scientific Literature. <em data-start=\"9299\" data-end=\"9344\">Journal of Data and Information Quality, 12<\/em>(3), 1\u201320.<\/p>\n<\/li>\n<\/ol>\n<h3>R\u00e9f\u00e9rences<\/h3>\n<ul>\n<li>Moreau et al., 2023, Semantic Scholar.<\/li>\n<li>Journal of Scientific Discovery, 2022.<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Mod\u00e8les hybrides NLP-graphes pour la d\u00e9couverte scientifique Auteur(s) : Dr. Jean Moreau \u2014 Date : 2023-07-02 \u2014 Source : Semantic Scholar R\u00e9sum\u00e9 (Abstract) L\u2019explosion de la production scientifique rend la d\u00e9couverte de connaissances pertinentes de plus en plus complexe. Les mod\u00e8les hybrides combinant le traitement du langage naturel (NLP) et les graphes de connaissances offrent [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":6287,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_bbp_topic_count":0,"_bbp_reply_count":0,"_bbp_total_topic_count":0,"_bbp_total_reply_count":0,"_bbp_voice_count":0,"_bbp_anonymous_reply_count":0,"_bbp_topic_count_hidden":0,"_bbp_reply_count_hidden":0,"_bbp_forum_subforum_count":0,"footnotes":""},"categories":[108],"tags":[],"class_list":["post-6248","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-informatique-intelligence-artificielle"],"acf":[],"_links":{"self":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/posts\/6248","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"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=6248"}],"version-history":[{"count":1,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/posts\/6248\/revisions"}],"predecessor-version":[{"id":6289,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/posts\/6248\/revisions\/6289"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/media\/6287"}],"wp:attachment":[{"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/media?parent=6248"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/categories?post=6248"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/sahelib.atatec-design.com\/index.php\/wp-json\/wp\/v2\/tags?post=6248"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}