{"id":453,"date":"2018-09-03T07:01:21","date_gmt":"2018-09-03T07:01:21","guid":{"rendered":"https:\/\/blog.feed.biz\/fr\/?p=453"},"modified":"2018-09-03T10:34:23","modified_gmt":"2018-09-03T10:34:23","slug":"classification-e-commerce-via-la-learning-machine","status":"publish","type":"post","link":"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/","title":{"rendered":"Classification e-commerce via la learning machine"},"content":{"rendered":"

Common-Services d\u00e9veloppe pour Feed.biz<\/a> et son offre conjointe avec Cdiscount le module Cdiscount Flux pour Prestashop<\/a>.<\/p>\n

Le probl\u00e9matique c’est que Cdiscount dispose de presque 15 000 cat\u00e9gories possibles de produits \u00e0 ce jour et a r\u00e9f\u00e9renc\u00e9 depuis sa cr\u00e9ation 44 millions de produits.<\/p>\n

La difficult\u00e9 pour le marchand est donc de pouvoir classifier ses produits correctement dans ces nombreuses cat\u00e9gories, quand un marchand dispose d’un nombre important de cat\u00e9gories cela peut devenir le parcours du combatant !<\/p>\n

Le challenge \u00e9tait donc pour nous initialement de r\u00e9aliser une cat\u00e9gorisation automatique, mais du fait que Cdiscount ne communique pas les descriptions des produits via les appels API qui permettent de consulter le catalogue, nous n’avons donc pu ne collecter que les noms de produits, ce qui fait un volume d’information plus succin, environ 4 Go de donn\u00e9es.<\/p>\n

Nous avons donc eu une approche diff\u00e9rente mais d\u00e9cid\u00e9 de mod\u00e9liser au lieu de cat\u00e9goriser; ce qui permet au marchand de se voir proposer des mod\u00e8les g\u00e9n\u00e9riques correspondant \u00e0 ses produits ; une liste probables de cat\u00e9gories, le genre du produit, le type de public, les attributs couleurs \u00e0 envoyer, si la cr\u00e9ation de variantes\/d\u00e9clinaisons est n\u00e9cessaire etc.<\/p>\n

Pour ce faire nous avons mis en oeuvre les diff\u00e9rents frameworks disponibles ; TensortFlow, Keras, Nltk, CnnText, FastText.<\/p>\n

La m\u00e9thodologie est assez simple en soit, il a suffit de normaliser, donc de rendre un texte complexe pour la machine en un texte intelligible, d’une part les cat\u00e9gories en utilisant le code cat\u00e9gorie comme cl\u00e9 et le nom 44 millions de produits classifi\u00e9s. La proc\u00e9dure de training pour l’ensemble de ces 44 millions de produits a dur\u00e9e 5 heures sur une machine disposant de 32 CPU cadenc\u00e9s \u00e0 2.7Ghz et disposant de 64 Go de RAM + 64 Go de RAM et d’un disque de 400 Go pour tampon.<\/p>\n

Il a fallut r\u00e9p\u00e9ter cette op\u00e9ration de multiples fois pour corriger tous les petits d\u00e9tails pouvant poser des probl\u00e8mes, notamment la ponctuation.<\/p>\n

Une fois le r\u00e9sultat parvenu, il \u00e9tait plut\u00f4t satisfaisant, nous avons obtenus des r\u00e9sultat pouvant aller jusqu’\u00e0 95% de pr\u00e9dictions justes ;<\/p>\n

Un exemple ici ;<\/p>\n

\"\"<\/a><\/p>\n

A gauche nous avons la cat\u00e9gorie du marchand qui n’est pas soumise \u00e0 la learning machine. A droite la cat\u00e9gorie correspondant \u00e0 Cdiscount. La donn\u00e9e soumise est le contenu int\u00e9gral des produits de la cat\u00e9gories du marchant, sans nommer la cat\u00e9gorie, car celle ci pourrait s’appeler par exemple accueil ! Nous voyons que l’int\u00e9gralit\u00e9 des produits soumis ont re\u00e7us comme suggestion les cat\u00e9gories Cdiscount les plus pertinentes.<\/p>\n

Une fois ce r\u00e9sultat obtenu, nous avons ainsi pu cr\u00e9er les mod\u00e8les probables correspondants ;<\/p>\n

\"\"<\/a><\/p>\n

Ainsi, d\u00e8s la rentr\u00e9e septembre le marchand qui soumettra un ensemble de cat\u00e9gories se verra proposer les mod\u00e8les les plus pertinents, il n’aura plus qu’a choisir parmi quelques uns plut\u00f4t que d’avoir \u00e0 r\u00e9aliser un mapping laborieux sur 15 000 cat\u00e9gories !<\/p>\n

Nous avons fait de m\u00eame pour les champs tailles et couleurs, pour les champs couleurs, nous avons entra\u00een\u00e9 un mod\u00e8le sur la base de la n\u00e9gation ; nous avons d’une part inventori\u00e9 toutes les couleurs et variantes de couleurs, puis, ajout\u00e9 au mod\u00e8le les n\u00e9gations les + \u00e9videntes, un maximum d’oppositions ; avion, chien, train, assiette etc… le proc\u00e9d\u00e9 est d’une simplicit\u00e9 enfantine, ne n\u00e9cessite par de lourds d\u00e9veloppements.<\/p>\n

Le temps n\u00e9cessaire pour la mod\u00e9lisation quand \u00e0 lui a n\u00e9cessit\u00e9 plus de 600 heures de travail !<\/p>\n

En pr\u00e9-production, pour exemple ici un marchand de chaussures, pour la cat\u00e9gorie « Sandales » ;<\/p>\n

\"\"<\/a><\/p>\n

Nous obtenons les pr\u00e9dictions suivantes ;<\/p>\n

\"\"<\/a><\/p>\n

 <\/p>\n

 <\/p>\n

Merci aux contributeurs des solutions mentionn\u00e9es ;<\/p>\n

TensortFlow ;\u00a0https:\/\/github.com\/tensorflow\/tensorflow<\/a>
\nKeras ; https:\/\/keras.io\/<\/a>
\nNltk ; https:\/\/www.nltk.org\/<\/a>
\nCnnText ; https:\/\/github.com\/dennybritz\/cnn-text-classification-tf<\/a>
\nFastText ;\u00a0https:\/\/github.com\/facebookresearch\/fastText\/<\/a><\/p>\n

Et \u00e0 merci \u00e0\u00a0Cdiscount<\/a>, Cdiscount Flux<\/a> pour leur aide pr\u00e9cieuse<\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n","protected":false},"excerpt":{"rendered":"

Common-Services d\u00e9veloppe pour Feed.biz et son offre conjointe avec Cdiscount le module Cdiscount Flux pour Prestashop. Le probl\u00e9matique c’est que Cdiscount dispose de presque 15 000 cat\u00e9gories possibles de produits \u00e0 ce jour et a r\u00e9f\u00e9renc\u00e9 depuis sa cr\u00e9ation 44 … Lire la suite\u00ad\u00ad<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31],"tags":[],"class_list":["post-453","post","type-post","status-publish","format-standard","hentry","category-cdiscount-fr"],"yoast_head":"\nClassification e-commerce via la learning machine - Feed.Biz<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Classification e-commerce via la learning machine - Feed.Biz\" \/>\n<meta property=\"og:description\" content=\"Common-Services d\u00e9veloppe pour Feed.biz et son offre conjointe avec Cdiscount le module Cdiscount Flux pour Prestashop. Le probl\u00e9matique c’est que Cdiscount dispose de presque 15 000 cat\u00e9gories possibles de produits \u00e0 ce jour et a r\u00e9f\u00e9renc\u00e9 depuis sa cr\u00e9ation 44 … Lire la suite\u00ad\u00ad\" \/>\n<meta property=\"og:url\" content=\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/\" \/>\n<meta property=\"og:site_name\" content=\"Feed.Biz\" \/>\n<meta property=\"article:published_time\" content=\"2018-09-03T07:01:21+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2018-09-03T10:34:23+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png\" \/>\n<meta name=\"author\" content=\"donotdont\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"donotdont\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/\"},\"author\":{\"name\":\"donotdont\",\"@id\":\"https:\/\/blog.feed.biz\/#\/schema\/person\/526f91821cf34bb472478b8a96c4462b\"},\"headline\":\"Classification e-commerce via la learning machine\",\"datePublished\":\"2018-09-03T07:01:21+00:00\",\"dateModified\":\"2018-09-03T10:34:23+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/\"},\"wordCount\":674,\"publisher\":{\"@id\":\"https:\/\/blog.feed.biz\/#organization\"},\"image\":{\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png\",\"articleSection\":[\"Cdiscount\"],\"inLanguage\":\"fr-FR\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/\",\"url\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/\",\"name\":\"Classification e-commerce via la learning machine - Feed.Biz\",\"isPartOf\":{\"@id\":\"https:\/\/blog.feed.biz\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png\",\"datePublished\":\"2018-09-03T07:01:21+00:00\",\"dateModified\":\"2018-09-03T10:34:23+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage\",\"url\":\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png\",\"contentUrl\":\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/blog.feed.biz\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Classification e-commerce via la learning machine\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/blog.feed.biz\/#website\",\"url\":\"https:\/\/blog.feed.biz\/\",\"name\":\"Feed.Biz\",\"description\":\"Feed.Biz\",\"publisher\":{\"@id\":\"https:\/\/blog.feed.biz\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/blog.feed.biz\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/blog.feed.biz\/#organization\",\"name\":\"Feed.biz\",\"url\":\"https:\/\/blog.feed.biz\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/blog.feed.biz\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2016\/09\/logo-dark.png\",\"contentUrl\":\"https:\/\/blog.feed.biz\/wp-content\/uploads\/2016\/09\/logo-dark.png\",\"width\":112,\"height\":22,\"caption\":\"Feed.biz\"},\"image\":{\"@id\":\"https:\/\/blog.feed.biz\/#\/schema\/logo\/image\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\/\/blog.feed.biz\/#\/schema\/person\/526f91821cf34bb472478b8a96c4462b\",\"name\":\"donotdont\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/blog.feed.biz\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/a063d6ca75f7fa7b33f65a63e9340b34?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/a063d6ca75f7fa7b33f65a63e9340b34?s=96&d=mm&r=g\",\"caption\":\"donotdont\"},\"url\":\"https:\/\/blog.feed.biz\/fr\/author\/donotdont\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Classification e-commerce via la learning machine - Feed.Biz","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/","og_locale":"fr_FR","og_type":"article","og_title":"Classification e-commerce via la learning machine - Feed.Biz","og_description":"Common-Services d\u00e9veloppe pour Feed.biz et son offre conjointe avec Cdiscount le module Cdiscount Flux pour Prestashop. Le probl\u00e9matique c’est que Cdiscount dispose de presque 15 000 cat\u00e9gories possibles de produits \u00e0 ce jour et a r\u00e9f\u00e9renc\u00e9 depuis sa cr\u00e9ation 44 … Lire la suite\u00ad\u00ad","og_url":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/","og_site_name":"Feed.Biz","article_published_time":"2018-09-03T07:01:21+00:00","article_modified_time":"2018-09-03T10:34:23+00:00","og_image":[{"url":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png"}],"author":"donotdont","twitter_misc":{"\u00c9crit par":"donotdont","Dur\u00e9e de lecture estim\u00e9e":"3 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#article","isPartOf":{"@id":"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/"},"author":{"name":"donotdont","@id":"https:\/\/blog.feed.biz\/#\/schema\/person\/526f91821cf34bb472478b8a96c4462b"},"headline":"Classification e-commerce via la learning machine","datePublished":"2018-09-03T07:01:21+00:00","dateModified":"2018-09-03T10:34:23+00:00","mainEntityOfPage":{"@id":"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/"},"wordCount":674,"publisher":{"@id":"https:\/\/blog.feed.biz\/#organization"},"image":{"@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage"},"thumbnailUrl":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png","articleSection":["Cdiscount"],"inLanguage":"fr-FR"},{"@type":"WebPage","@id":"https:\/\/blog.feed.biz\/fr\/classification-e-commerce-via-la-learning-machine\/","url":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/","name":"Classification e-commerce via la learning machine - Feed.Biz","isPartOf":{"@id":"https:\/\/blog.feed.biz\/#website"},"primaryImageOfPage":{"@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage"},"image":{"@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage"},"thumbnailUrl":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png","datePublished":"2018-09-03T07:01:21+00:00","dateModified":"2018-09-03T10:34:23+00:00","breadcrumb":{"@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#primaryimage","url":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png","contentUrl":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2018\/09\/products-classification-by-learning-machine01.png"},{"@type":"BreadcrumbList","@id":"https:\/\/blog.common-services.com\/classification-e-commerce-via-la-learning-machine\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/blog.feed.biz\/"},{"@type":"ListItem","position":2,"name":"Classification e-commerce via la learning machine"}]},{"@type":"WebSite","@id":"https:\/\/blog.feed.biz\/#website","url":"https:\/\/blog.feed.biz\/","name":"Feed.Biz","description":"Feed.Biz","publisher":{"@id":"https:\/\/blog.feed.biz\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/blog.feed.biz\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-FR"},{"@type":"Organization","@id":"https:\/\/blog.feed.biz\/#organization","name":"Feed.biz","url":"https:\/\/blog.feed.biz\/","logo":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/blog.feed.biz\/#\/schema\/logo\/image\/","url":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2016\/09\/logo-dark.png","contentUrl":"https:\/\/blog.feed.biz\/wp-content\/uploads\/2016\/09\/logo-dark.png","width":112,"height":22,"caption":"Feed.biz"},"image":{"@id":"https:\/\/blog.feed.biz\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/blog.feed.biz\/#\/schema\/person\/526f91821cf34bb472478b8a96c4462b","name":"donotdont","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/blog.feed.biz\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/a063d6ca75f7fa7b33f65a63e9340b34?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/a063d6ca75f7fa7b33f65a63e9340b34?s=96&d=mm&r=g","caption":"donotdont"},"url":"https:\/\/blog.feed.biz\/fr\/author\/donotdont\/"}]}},"_links":{"self":[{"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/posts\/453"}],"collection":[{"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/comments?post=453"}],"version-history":[{"count":5,"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/posts\/453\/revisions"}],"predecessor-version":[{"id":465,"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/posts\/453\/revisions\/465"}],"wp:attachment":[{"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/media?parent=453"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/categories?post=453"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.feed.biz\/fr\/wp-json\/wp\/v2\/tags?post=453"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}