Il existe toutefois un probl\u00e8me central avec RAG : les solutions traditionnelles ignorent le contexte global, ce qui conduit souvent le syst\u00e8me \u00e0 ne pas trouver d\u2019informations pertinentes dans la base de connaissances. Par exemple, un utilisateur hypoth\u00e9tique pourrait poser la question suivante : \u00ab Quelles d\u00e9cisions ont \u00e9t\u00e9 prises concernant les prochaines \u00e9tapes \u00e0 l\u2019\u00e9cole Max Mustermann ? \u00bb Cet utilisateur souhaite obtenir des informations sur les d\u00e9lais, les \u00e9tapes \u00e0 venir et les accords pr\u00e9vus dans le cadre de la construction de l\u2019\u00e9cole. Pourtant, le chunk correspondant, qui contient la r\u00e9ponse, peut ne pas comporter de r\u00e9f\u00e9rence explicite \u00e0 l\u2019\u00e9cole Max Mustermann et ne peut donc pas \u00eatre correctement r\u00e9cup\u00e9r\u00e9 pour g\u00e9n\u00e9rer la r\u00e9ponse.<\/p>\n
![\"L\u2019image](\"https:\/\/www.charlyverse.com\/wp-content\/uploads\/2025\/10\/Picture1.png\")
<\/p>\n
Dans mon m\u00e9moire de master, j\u2019ai \u00e9tudi\u00e9 ce probl\u00e8me et mis en \u0153uvre l\u2019approche \u00ab <\/strong>Contextual-RAG \u00bb<\/strong> avec le corpus des protocoles parlementaires du groupe de recherche DisLab de l\u2019universit\u00e9 technique de Darmstadt. L\u2019\u00e9tape de contextualisation a lieu apr\u00e8s la division du texte en chunks et avant leur conversion en embeddings. Chaque chunk est enrichi d\u2019un contexte pertinent, issu d\u2019\u00e9nonc\u00e9s pr\u00e9c\u00e9dents ou suivants, ou m\u00eame li\u00e9 \u00e0 l\u2019id\u00e9e principale de l\u2019ensemble du texte. Chaque chunk est ainsi compl\u00e9t\u00e9 par des indices s\u00e9mantiques qui n\u2019\u00e9taient pas pr\u00e9sents auparavant mais qui peuvent d\u00e9sormais correspondre \u00e0 une requ\u00eate d\u2019utilisateur. Cette contextualisation a \u00e9t\u00e9 r\u00e9alis\u00e9e \u00e0 l\u2019aide de mod\u00e8les linguistiques locaux disponibles publiquement. Les r\u00e9sultats permettent de conclure que l\u2019application de Contextual-RAG \u00e0 la recherche dans un corpus politique atteint jusqu\u2019\u00e0 18 % de meilleures performances de rappel que l\u2019utilisation de Basic-RAG. Cela signifie que davantage de r\u00e9sultats pertinents sont trouv\u00e9s parmi les top-k r\u00e9ponses.<\/p>\n Les connaissances et m\u00e9thodes acquises dans ce travail seront directement int\u00e9gr\u00e9es au logiciel d\u2019entreprise Inno:docs. Inno:docs est un chatbot bas\u00e9 sur l\u2019IA qui permet de rechercher des donn\u00e9es d\u2019entreprise en langage naturel et fournit des r\u00e9ponses factuelles. L\u2019approche Contextual-RAG permet au syst\u00e8me de trouver davantage de documents pertinents, puisque chaque chunk prend en compte le contexte complet du texte et n\u2019est pas stock\u00e9 isol\u00e9ment dans la base de donn\u00e9es.<\/p>\n <\/p>\n Image : Innomatik AG<\/em><\/p>\n <\/p>\n","protected":false},"excerpt":{"rendered":" Nous sommes tr\u00e8s heureux de partager aujourd\u2019hui une histoire de r\u00e9ussite toute particuli\u00e8re de notre \u00e9quipe : notre coll\u00e8gue Ernesto Curmen a \u00e9t\u00e9 distingu\u00e9 pour son m\u00e9moire de master exceptionnel. Dans sa contribution, il nous donne lui-m\u00eame un aper\u00e7u du contenu passionnant et raconte quels th\u00e8mes l\u2019ont particuli\u00e8rement marqu\u00e9. Bonne lecture! Retrieval Augmented Generation, ou […]<\/p>\n","protected":false},"author":8,"featured_media":2518,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ocean_post_layout":"","ocean_both_sidebars_style":"","ocean_both_sidebars_content_width":0,"ocean_both_sidebars_sidebars_width":0,"ocean_sidebar":"0","ocean_second_sidebar":"0","ocean_disable_margins":"enable","ocean_add_body_class":"","ocean_shortcode_before_top_bar":"","ocean_shortcode_after_top_bar":"","ocean_shortcode_before_header":"","ocean_shortcode_after_header":"","ocean_has_shortcode":"","ocean_shortcode_after_title":"","ocean_shortcode_before_footer_widgets":"","ocean_shortcode_after_footer_widgets":"","ocean_shortcode_before_footer_bottom":"","ocean_shortcode_after_footer_bottom":"","ocean_display_top_bar":"default","ocean_display_header":"default","ocean_header_style":"","ocean_center_header_left_menu":"0","ocean_custom_header_template":"0","ocean_custom_logo":0,"ocean_custom_retina_logo":0,"ocean_custom_logo_max_width":0,"ocean_custom_logo_tablet_max_width":0,"ocean_custom_logo_mobile_max_width":0,"ocean_custom_logo_max_height":0,"ocean_custom_logo_tablet_max_height":0,"ocean_custom_logo_mobile_max_height":0,"ocean_header_custom_menu":"0","ocean_menu_typo_font_family":"0","ocean_menu_typo_font_subset":"","ocean_menu_typo_font_size":0,"ocean_menu_typo_font_size_tablet":0,"ocean_menu_typo_font_size_mobile":0,"ocean_menu_typo_font_size_unit":"px","ocean_menu_typo_font_weight":"","ocean_menu_typo_font_weight_tablet":"","ocean_menu_typo_font_weight_mobile":"","ocean_menu_typo_transform":"","ocean_menu_typo_transform_tablet":"","ocean_menu_typo_transform_mobile":"","ocean_menu_typo_line_height":0,"ocean_menu_typo_line_height_tablet":0,"ocean_menu_typo_line_height_mobile":0,"ocean_menu_typo_line_height_unit":"","ocean_menu_typo_spacing":0,"ocean_menu_typo_spacing_tablet":0,"ocean_menu_typo_spacing_mobile":0,"ocean_menu_typo_spacing_unit":"","ocean_menu_link_color":"","ocean_menu_link_color_hover":"","ocean_menu_link_color_active":"","ocean_menu_link_background":"","ocean_menu_link_hover_background":"","ocean_menu_link_active_background":"","ocean_menu_social_links_bg":"","ocean_menu_social_hover_links_bg":"","ocean_menu_social_links_color":"","ocean_menu_social_hover_links_color":"","ocean_disable_title":"default","ocean_disable_heading":"default","ocean_post_title":"","ocean_post_subheading":"","ocean_post_title_style":"","ocean_post_title_background_color":"","ocean_post_title_background":0,"ocean_post_title_bg_image_position":"","ocean_post_title_bg_image_attachment":"","ocean_post_title_bg_image_repeat":"","ocean_post_title_bg_image_size":"","ocean_post_title_height":0,"ocean_post_title_bg_overlay":0.5,"ocean_post_title_bg_overlay_color":"","ocean_disable_breadcrumbs":"default","ocean_breadcrumbs_color":"","ocean_breadcrumbs_separator_color":"","ocean_breadcrumbs_links_color":"","ocean_breadcrumbs_links_hover_color":"","ocean_display_footer_widgets":"default","ocean_display_footer_bottom":"default","ocean_custom_footer_template":"0","osh_disable_topbar_sticky":"default","osh_disable_header_sticky":"default","osh_sticky_header_style":"default","osh_sticky_header_effect":"","osh_custom_sticky_logo":0,"osh_custom_retina_sticky_logo":0,"osh_custom_sticky_logo_height":0,"osh_background_color":"","osh_links_color":"","osh_links_hover_color":"","osh_links_active_color":"","osh_links_bg_color":"","osh_links_hover_bg_color":"","osh_links_active_bg_color":"","osh_menu_social_links_color":"","osh_menu_social_hover_links_color":"","ocean_post_oembed":"","ocean_post_self_hosted_media":"","ocean_post_video_embed":"","ocean_link_format":"","ocean_link_format_target":"self","ocean_quote_format":"","ocean_quote_format_link":"post","ocean_gallery_link_images":"on","ocean_gallery_id":[],"footnotes":""},"categories":[111],"tags":[236,246,178,429,437,436],"class_list":["post-2524","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ia-robotique","tag-forschung-fr","tag-ia","tag-ki-fr","tag-logiciel","tag-rag","tag-recherche","entry","has-media"],"yoast_head":"\n![\"L\u2019image](\"https:\/\/www.charlyverse.com\/wp-content\/uploads\/2025\/10\/Picture2.png\")
<\/p>\n