	{"id":65530,"date":"2021-12-13T11:26:57","date_gmt":"2021-12-13T11:26:57","guid":{"rendered":"https:\/\/www.artefact.com\/?post_type=news&#038;p=65530"},"modified":"2024-09-20T17:45:47","modified_gmt":"2024-09-20T16:45:47","slug":"scoring-customer-propensity-using-machine-learning-models-on-google-analytics-data","status":"publish","type":"blog","link":"https:\/\/www.artefact.com\/br\/blog\/scoring-customer-propensity-using-machine-learning-models-on-google-analytics-data\/","title":{"rendered":"Pontua\u00e7\u00e3o da propens\u00e3o do cliente usando modelos de aprendizado de m\u00e1quina no Google Analytics Data"},"content":{"rendered":"<p><div class=\"fusion-fullwidth fullwidth-box fusion-builder-row-1 fusion-flex-container nonhundred-percent-fullwidth non-hundred-percent-height-scrolling article-author\" style=\"--awb-border-radius-top-left:0px;--awb-border-radius-top-right:0px;--awb-border-radius-bottom-right:0px;--awb-border-radius-bottom-left:0px;--awb-background-color:#ffffff;--awb-flex-wrap:wrap;\" ><div class=\"fusion-builder-row fusion-row fusion-flex-align-items-flex-start fusion-flex-content-wrap\" style=\"max-width:calc( 1440px + 20px );margin-left: calc(-20px \/ 2 );margin-right: calc(-20px \/ 2 );\"><div class=\"fusion-layout-column fusion_builder_column fusion-builder-column-0 fusion_builder_column_1_2 1_2 fusion-flex-column\" style=\"--awb-bg-size:cover;--awb-width-large:50%;--awb-margin-top-large:0px;--awb-spacing-right-large:10px;--awb-margin-bottom-large:0px;--awb-spacing-left-large:10px;--awb-width-medium:50%;--awb-order-medium:0;--awb-spacing-right-medium:10px;--awb-spacing-left-medium:10px;--awb-width-small:100%;--awb-order-small:0;--awb-spacing-right-small:10px;--awb-spacing-left-small:10px;\"><div class=\"fusion-column-wrapper fusion-column-has-shadow fusion-flex-justify-content-flex-start fusion-content-layout-column\"><div class=\"fusion-title title fusion-title-1 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">Autor<\/h2><\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27150%27%20height%3D%270%27%20viewBox%3D%270%200%20150%200%27%3E%3Crect%20width%3D%27150%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/05\/Antoine-Aubay.jpeg\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left article-author-image\" style=\"width: 150px; border-radius: 54% 46% 77% 23% \/ 74% 40% 60% 26%; overflow: hidden;\" width=\"150\" height=\"auto\" \/><div class=\"fusion-title title fusion-title-2 fusion-sep-none fusion-title-text fusion-title-size-three article-author-name-title\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Antoine Aubay<\/h3><\/div><div class=\"fusion-text fusion-text-1 article-author-description\"><p>Cientista s\u00eanior do Data no Artefact<\/p>\n<\/div><\/div><\/div><\/div><\/div><div class=\"fusion-fullwidth fullwidth-box fusion-builder-row-2 fusion-flex-container nonhundred-percent-fullwidth non-hundred-percent-height-scrolling\" style=\"--awb-border-radius-top-left:0px;--awb-border-radius-top-right:0px;--awb-border-radius-bottom-right:0px;--awb-border-radius-bottom-left:0px;--awb-margin-top:40px;--awb-margin-bottom:40px;--awb-flex-wrap:wrap;\" ><div class=\"fusion-builder-row fusion-row fusion-flex-align-items-center fusion-flex-justify-content-center fusion-flex-content-wrap\" style=\"max-width:calc( 1440px + 20px );margin-left: calc(-20px \/ 2 );margin-right: calc(-20px \/ 2 );\"><div class=\"fusion-layout-column fusion_builder_column fusion-builder-column-1 fusion_builder_column_1_1 1_1 fusion-flex-column fusion-flex-align-self-center fusion-column-inner-bg-wrapper\" style=\"--awb-padding-top:20px;--awb-padding-right:20px;--awb-padding-bottom:20px;--awb-padding-left:20px;--awb-overflow:hidden;--awb-inner-bg-size:cover;--awb-border-color:rgba(10,17,40,0.1);--awb-border-top:1px;--awb-border-right:1px;--awb-border-bottom:1px;--awb-border-left:1px;--awb-border-style:solid;--awb-border-radius:4px 4px 4px 4px;--awb-inner-bg-border-radius:4px 4px 4px 4px;--awb-inner-bg-overflow:hidden;--awb-width-large:100%;--awb-margin-top-large:0px;--awb-spacing-right-large:10px;--awb-margin-bottom-large:0px;--awb-spacing-left-large:10px;--awb-width-medium:100%;--awb-order-medium:0;--awb-spacing-right-medium:10px;--awb-spacing-left-medium:10px;--awb-width-small:100%;--awb-order-small:0;--awb-spacing-right-small:10px;--awb-spacing-left-small:10px;\"><span class=\"fusion-column-inner-bg hover-type-none\"><a class=\"fusion-column-anchor\" href=\"https:\/\/medium.com\/artefact-engineering-and-data-science\/scoring-customer-propensity-using-machine-learning-models-on-google-analytics-data-ba1126469c1f\" rel=\"noopener noreferrer\" target=\"_blank\"><span class=\"fusion-column-inner-bg-image\"><\/span><\/a><\/span><div class=\"fusion-column-wrapper fusion-column-has-shadow fusion-flex-justify-content-center fusion-content-layout-row fusion-flex-align-items-center\"><div class=\"fusion-text fusion-text-2\"><p><u>Leia nosso artigo sobre<\/u><\/p>\n<\/div><div class=\"fusion-image-element\" style=\"--awb-margin-right:20px;--awb-margin-left:20px;--awb-max-width:150px;--awb-caption-title-font-family:var(--h2_typography-font-family);--awb-caption-title-font-weight:var(--h2_typography-font-weight);--awb-caption-title-font-style:var(--h2_typography-font-style);--awb-caption-title-size:var(--h2_typography-font-size);--awb-caption-title-transform:var(--h2_typography-text-transform);--awb-caption-title-line-height:var(--h2_typography-line-height);--awb-caption-title-letter-spacing:var(--h2_typography-letter-spacing);\"><span class=\"fusion-imageframe imageframe-none imageframe-1 hover-type-none\"><img decoding=\"async\" width=\"4000\" height=\"992\" title=\"M\u00e9dio Blog\" src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog.png\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog.png\" alt class=\"lazyload img-responsive wp-image-60582\" srcset=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%274000%27%20height%3D%27992%27%20viewBox%3D%270%200%204000%20992%27%3E%3Crect%20width%3D%274000%27%20height%3D%27992%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-srcset=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog-200x50.png 200w, https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog-400x99.png 400w, https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog-600x149.png 600w, https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog-800x198.png 800w, https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog-1200x298.png 1200w, https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/04\/Medium-Blog.png 4000w\" data-sizes=\"auto\" data-orig-sizes=\"(max-width: 640px) 100vw, 4000px\" \/><\/span><\/div><div class=\"fusion-text fusion-text-3\"><p>.<\/p>\n<\/div><\/div><\/div><\/div><\/div><div class=\"fusion-fullwidth fullwidth-box fusion-builder-row-3 fusion-flex-container nonhundred-percent-fullwidth non-hundred-percent-height-scrolling\" style=\"--awb-border-radius-top-left:0px;--awb-border-radius-top-right:0px;--awb-border-radius-bottom-right:0px;--awb-border-radius-bottom-left:0px;--awb-flex-wrap:wrap;\" ><div class=\"fusion-builder-row fusion-row fusion-flex-align-items-flex-start fusion-flex-content-wrap\" style=\"max-width:calc( 1440px + 20px );margin-left: calc(-20px \/ 2 );margin-right: calc(-20px \/ 2 );\"><div class=\"fusion-layout-column fusion_builder_column fusion-builder-column-2 fusion_builder_column_1_1 1_1 fusion-flex-column\" style=\"--awb-bg-size:cover;--awb-width-large:100%;--awb-margin-top-large:0px;--awb-spacing-right-large:10px;--awb-margin-bottom-large:0px;--awb-spacing-left-large:10px;--awb-width-medium:100%;--awb-order-medium:0;--awb-spacing-right-medium:10px;--awb-spacing-left-medium:10px;--awb-width-small:100%;--awb-order-small:0;--awb-spacing-right-small:10px;--awb-spacing-left-small:10px;\"><div class=\"fusion-column-wrapper fusion-column-has-shadow fusion-flex-justify-content-flex-start fusion-content-layout-column\"><div class=\"fusion-text fusion-text-4 description\"><p>A modelagem de propens\u00e3o pode ser usada para aumentar o impacto de sua comunica\u00e7\u00e3o com os clientes e otimizar os gastos com o or\u00e7amento de publicidade.<\/p>\n<p>O data do Google Analytics \u00e9 uma fonte data bem estruturada que pode ser facilmente transformada em um conjunto data pronto para aprendizado de m\u00e1quina.<\/p>\n<p>O backtest no hist\u00f3rico data e as m\u00e9tricas t\u00e9cnicas podem lhe dar uma primeira no\u00e7\u00e3o do desempenho do seu modelo, enquanto o teste ao vivo e as m\u00e9tricas comerciais lhe permitir\u00e3o confirmar o impacto do seu modelo.<\/p>\n<p>Nosso modelo personalizado de aprendizado de m\u00e1quina superou as linhas de base existentes: durante os testes ao vivo em termos de ROAS (Return on advertising spend): +221% vs. modelo baseado em regras e +73% vs. aprendizado de m\u00e1quina pronto para uso (\u00edndice de qualidade da sess\u00e3o do Google Analytics).<\/p>\n<p>Este artigo pressup\u00f5e fundamentos b\u00e1sicos de aprendizado de m\u00e1quina e marketing.<\/p>\n<\/div><\/div><\/div><\/div><\/div><article class=\"fusion-fullwidth fullwidth-box fusion-builder-row-4 fusion-flex-container nonhundred-percent-fullwidth non-hundred-percent-height-scrolling\" style=\"--awb-border-radius-top-left:0px;--awb-border-radius-top-right:0px;--awb-border-radius-bottom-right:0px;--awb-border-radius-bottom-left:0px;--awb-flex-wrap:wrap;\" ><div class=\"fusion-builder-row fusion-row fusion-flex-align-items-flex-start fusion-flex-justify-content-center fusion-flex-content-wrap\" style=\"max-width:calc( 1440px + 20px );margin-left: calc(-20px \/ 2 );margin-right: calc(-20px \/ 2 );\"><div class=\"fusion-layout-column fusion_builder_column fusion-builder-column-3 fusion_builder_column_1_1 1_1 fusion-flex-column\" style=\"--awb-bg-size:cover;--awb-width-large:100%;--awb-margin-top-large:0px;--awb-spacing-right-large:10px;--awb-margin-bottom-large:0px;--awb-spacing-left-large:10px;--awb-width-medium:100%;--awb-order-medium:0;--awb-spacing-right-medium:10px;--awb-spacing-left-medium:10px;--awb-width-small:100%;--awb-order-small:0;--awb-spacing-right-small:10px;--awb-spacing-left-small:10px;\"><div class=\"fusion-column-wrapper fusion-column-has-shadow fusion-flex-justify-content-flex-start fusion-content-layout-column\"><div class=\"fusion-title title fusion-title-3 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">O que \u00e9 modelagem de propens\u00e3o?<\/h2><\/div><div class=\"fusion-text fusion-text-5\"><p>A modelagem de propens\u00e3o \u00e9\u00a0<strong>estimar a probabilidade de um cliente realizar uma determinada a\u00e7\u00e3o<\/strong>. H\u00e1 v\u00e1rias a\u00e7\u00f5es que podem ser \u00fateis para estimar:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-1 fusion-checklist-default type-icons\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Compra de um produto<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Agita\u00e7\u00e3o<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Cancelamento da inscri\u00e7\u00e3o<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">\n<p>etc ...<\/p>\n<\/div><\/li><\/ul><div class=\"fusion-text fusion-text-6\"><p>Neste artigo, vamos nos concentrar em estimar a propens\u00e3o de comprar um item em um site de com\u00e9rcio eletr\u00f4nico.<\/p>\n<p>Mas por que estimar a propens\u00e3o a comprar? Porque isso permite<strong>adaptar a forma como queremos interagir com um cliente.<\/strong> Por exemplo, suponha que tenhamos um modelo de propens\u00e3o muito simples que classifica os clientes em \u201cFrios\u201d, \u201cMornos\u201d e \u201cQuentes\u201d para um determinado produto (\u201cQuentes\u201d s\u00e3o os clientes com maior chance de compra e \u201cFrios\u201d, a menor):<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27619%27%20height%3D%270%27%20viewBox%3D%270%200%20619%200%27%3E%3Crect%20width%3D%27619%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article1.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 619px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"619\" height=\"auto\" \/><div class=\"fusion-text fusion-text-7\"><p>Bem, com base nessa classifica\u00e7\u00e3o<strong>o senhor pode ter uma resposta direcionada espec\u00edfica para cada classe<\/strong>. Talvez o senhor queira ter uma abordagem de marketing diferente com um cliente que est\u00e1 muito pr\u00f3ximo de comprar do que com um que talvez nem tenha ouvido falar do seu produto. Al\u00e9m disso, se o senhor tiver um or\u00e7amento de m\u00eddia limitado, poder\u00e1 concentr\u00e1-lo nos clientes que t\u00eam grande probabilidade de comprar e n\u00e3o gastar muito com aqueles que t\u00eam poucas chances.<\/p>\n<p>Esse tipo simples de classifica\u00e7\u00e3o baseada em regras pode dar bons resultados e, em geral, \u00e9 melhor do que n\u00e3o ter nenhuma, mas tem <strong>v\u00e1rias limita\u00e7\u00f5es:<\/strong><\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-2 fusion-checklist-default type-icons paddingList dark-text\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">\n<p>\u00c9 prov\u00e1vel que\u00a0<strong>n\u00e3o explorando todos os data <\/strong>que o senhor tem \u00e0 sua disposi\u00e7\u00e3o, sejam informa\u00e7\u00f5es mais precisas sobre a jornada do cliente, seu site ou outras fontes data que o senhor possa ter \u00e0 sua disposi\u00e7\u00e3o, como o CRM data.<\/p>\n<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Embora pare\u00e7a \u00f3bvio que os clientes classificados como \u201cquentes\u201d t\u00eam maior probabilidade de comprar do que os \u201cmornos\u201d, que t\u00eam maior probabilidade de comprar do que os \u201cfrios\u201d, essa abordagem n\u00e3o nos fornece n\u00fameros espec\u00edficos sobre <strong>qual a probabilidade de compra<\/strong>. Os clientes \u201cquentes\u201d t\u00eam 3% chance de comprar? 5%? 10% ?<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Usando regras simples, o n\u00famero de classes que o senhor pode obter \u00e9 limitado, o que\u00a0<strong>limita a personaliza\u00e7\u00e3o de sua resposta direcionada<\/strong>\u00a0pode ser.<\/div><\/li><\/ul><div class=\"fusion-text fusion-text-8\"><p>Para lidar com essas limita\u00e7\u00f5es, podemos usar uma abordagem mais voltada para o data: usar\u00a0<strong>aprendizado de m\u00e1quina <\/strong>em nosso data para\u00a0<strong>prever uma probabilidade de compra<\/strong>\u00a0para cada cliente.<\/p>\n<\/div><div class=\"fusion-title title fusion-title-4 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">Entendendo o Google Analytics data<\/h2><\/div><div class=\"fusion-text fusion-text-9\"><p>O Google Analytics \u00e9 um\u00a0<strong>servi\u00e7o anal\u00edtico da web <\/strong>que rastreia o uso data e o tr\u00e1fego em sites e aplicativos.<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27587%27%20height%3D%270%27%20viewBox%3D%270%200%20587%200%27%3E%3Crect%20width%3D%27587%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article2.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 587px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"587\" height=\"auto\" \/><div class=\"fusion-text fusion-text-10\"><p><a class=\"bv kv\" href=\"https:\/\/support.google.com\/analytics\/answer\/3416092?hl=en#zippy=%2Cin-this-article\" target=\"_blank\" rel=\"noopener ugc nofollow\">O Google Analytics data pode ser<strong>\u00a0facilmente exportado para o Big Query<\/strong>\u00a0<\/a>(Google Cloud Platform totalmente gerenciado\u00a0<a class=\"bv kv\" href=\"https:\/\/cloud.google.com\/bigquery\/?utm_source=google&amp;utm_medium=cpc&amp;utm_campaign=emea-gb-all-en-dr-skws-all-solutions-trial-b-gcp-1010042&amp;utm_content=text-ad-none-any-DEV_c-CRE_335630920539-ADGP_Hybrid+%7C+SKWS+-+BMM+%7C+Txt+~+Data+Analytics+~+BigQuery%23v1-KWID_43700053279032269-kwd-47616964923-userloc_1006094&amp;utm_term=KW_%2Bbigquery-NET_g-PLAC_&amp;gclid=CjwKCAjwiY6MBhBqEiwARFSCPqzx1ubPaHp-g3MMEY8zES0fgiSrD3RYgcBGjQeNRRcV_EiS10fZ_RoCUgcQAvD_BwE&amp;gclsrc=aw.ds\" target=\"_blank\" rel=\"noopener ugc nofollow\">Servi\u00e7o de armaz\u00e9m data<\/a>), onde ele pode ser acessado por meio de uma sintaxe semelhante \u00e0 do SQL:<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27598%27%20height%3D%270%27%20viewBox%3D%270%200%20598%200%27%3E%3Crect%20width%3D%27598%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/article-aubay3.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 598px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"598\" height=\"auto\" \/><div class=\"fusion-text fusion-text-11\"><p>Observe que a tabela de exporta\u00e7\u00e3o do Big Query com o Google Analytics data \u00e9 um <strong>tabela aninhada no n\u00edvel da sess\u00e3o:<\/strong><\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-3 fusion-checklist-default type-icons\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\"><a class=\"bv kv\" href=\"https:\/\/support.google.com\/analytics\/answer\/2731565?hl=en#zippy=%2Cin-this-article\" target=\"_blank\" rel=\"noopener ugc nofollow\">Sess\u00f5es<\/a>\u00a0s\u00e3o uma lista de a\u00e7\u00f5es que um cliente espec\u00edfico realiza em um determinado per\u00edodo de tempo. Elas come\u00e7am quando um cliente visita uma p\u00e1gina e terminam ap\u00f3s 30 minutos de atividade.<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Cada cliente pode ter v\u00e1rias sess\u00f5es.<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Cada sess\u00e3o pode ser composta de v\u00e1rias ocorr\u00eancias (ou seja, eventos) e cada ocorr\u00eancia pode ter v\u00e1rios atributos ou m\u00e9tricas personalizadas (\u00e9 por isso que a tabela \u00e9 aninhada; por exemplo, se o senhor quiser examinar o data no n\u00edvel da ocorr\u00eancia, precisar\u00e1 achatar a tabela).<\/div><\/li><\/ul><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27619%27%20height%3D%270%27%20viewBox%3D%270%200%20619%200%27%3E%3Crect%20width%3D%27619%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article4.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 619px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"619\" height=\"auto\" \/><div class=\"fusion-text fusion-text-12\"><p>Por exemplo, nesta consulta, estamos analisando apenas\u00a0<strong>recursos em n\u00edvel de sess\u00e3o<\/strong>:<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27564%27%20height%3D%270%27%20viewBox%3D%270%200%20564%200%27%3E%3Crect%20width%3D%27564%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article5.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 564px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"564\" height=\"auto\" \/><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27686%27%20height%3D%270%27%20viewBox%3D%270%200%20686%200%27%3E%3Crect%20width%3D%27686%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article6.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 686px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"686\" height=\"auto\" \/><div class=\"fusion-text fusion-text-13\"><p>E, nessa consulta, usamos uma fun\u00e7\u00e3o Unnest para consultar as mesmas informa\u00e7\u00f5es em <strong>n\u00edvel de acerto<\/strong>:<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27686%27%20height%3D%270%27%20viewBox%3D%270%200%20686%200%27%3E%3Crect%20width%3D%27686%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article7.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 686px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"686\" height=\"auto\" \/><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27686%27%20height%3D%270%27%20viewBox%3D%270%200%20686%200%27%3E%3Crect%20width%3D%27686%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article8.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 686px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"686\" height=\"auto\" \/><div class=\"fusion-text fusion-text-14\"><p>Para obter mais informa\u00e7\u00f5es sobre o GA data, consulte o\u00a0<a class=\"bv kv\" href=\"https:\/\/developers.google.com\/analytics\" target=\"_blank\" rel=\"noopener ugc nofollow\">documenta\u00e7\u00e3o<\/a>. Observe que nosso projeto foi desenvolvido no GA360, portanto, se o senhor estiver usando a vers\u00e3o mais recente, GA4, haver\u00e1 algumas pequenas diferen\u00e7as no modelo data, especialmente a tabela que estar\u00e1 no n\u00edvel do evento. H\u00e1 tabelas de amostra p\u00fablicas do\u00a0<a class=\"bv kv\" href=\"https:\/\/console.cloud.google.com\/marketplace\/product\/obfuscated-ga360-data\/obfuscated-ga360-data?project=lexical-script-761\" target=\"_blank\" rel=\"noopener ugc nofollow\">GA360<\/a>\u00a0e\u00a0<a class=\"bv kv\" href=\"https:\/\/support.google.com\/analytics\/answer\/10937659?hl=en&amp;ref_topic=9359001#zippy=%2Cin-this-article\" target=\"_blank\" rel=\"noopener ugc nofollow\">GA4<\/a>\u00a0data dispon\u00edvel no Big Query.<\/p>\n<p>Agora que temos acesso \u00e0 nossa fonte bruta data, precisamos realizar a engenharia de recursos antes de podermos alimentar nossa tabela com um algoritmo de aprendizado de m\u00e1quina<\/p>\n<\/div><div class=\"fusion-title title fusion-title-5 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">Criando os recursos certos<\/h2><\/div><div class=\"fusion-text fusion-text-15\"><p>O objetivo da etapa de engenharia de recursos \u00e9 transformar o data bruto do Google Analytics (extra\u00eddo do Big Query) em um\u00a0<strong>mesa pronta <\/strong>a ser usado para<strong>Aprendizado de m\u00e1quina<\/strong>.<\/p>\n<p>O GA data \u00e9 muito bem estruturado e exigir\u00e1 o m\u00ednimo de etapas de limpeza data. No entanto, ainda h\u00e1 muitas informa\u00e7\u00f5es presentes na tabela, muitas das quais n\u00e3o s\u00e3o \u00fateis para o aprendizado de m\u00e1quina ou n\u00e3o podem ser usadas como est\u00e3o, portanto, \u00e9 importante selecionar e criar os recursos certos. Para isso, criamos recursos que pareciam ser os mais correlacionados com a compra de um produto.<\/p>\n<p>Criamos quatro tipos de recursos:<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27625%27%20height%3D%270%27%20viewBox%3D%270%200%20625%200%27%3E%3Crect%20width%3D%27625%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article9.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 625px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"625\" height=\"auto\" \/><div class=\"fusion-text fusion-text-16\"><p>Observe que estamos computando todos esses recursos no n\u00edvel do cliente, o que significa que estamos agregando informa\u00e7\u00f5es de v\u00e1rias sess\u00f5es para cada cliente (usando o campo fullVisitorId como chave)<\/p>\n<\/div><div class=\"fusion-title title fusion-title-6 fusion-sep-none fusion-title-text fusion-title-size-three\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Caracter\u00edsticas gerais<\/h3><\/div><div class=\"fusion-text fusion-text-17\"><p>Os recursos globais s\u00e3o\u00a0<strong>caracter\u00edsticas num\u00e9ricas<\/strong>\u00a0que fornecem informa\u00e7\u00f5es gerais sobre a sess\u00e3o.<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27524%27%20height%3D%270%27%20viewBox%3D%270%200%20524%200%27%3E%3Crect%20width%3D%27524%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article10.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 524px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"524\" height=\"auto\" \/><div class=\"fusion-text fusion-text-18\"><p>Observe que a taxa de rejei\u00e7\u00e3o \u00e9 definida como % de vezes que o cliente visitou apenas uma p\u00e1gina da Web durante uma sess\u00e3o.<\/p>\n<p>Tamb\u00e9m foi importante incluir informa\u00e7\u00f5es sobre o <strong>rec\u00eancia de eventos<\/strong>Por exemplo, um cliente que acabou de visitar seu site provavelmente est\u00e1 mais disposto a comprar do que um que o visitou h\u00e1 tr\u00eas meses. Para obter mais informa\u00e7\u00f5es sobre esse t\u00f3pico, o senhor pode consultar a teoria em\u00a0<a class=\"bv kv\" href=\"https:\/\/en.wikipedia.org\/wiki\/RFM_(market_research)\" target=\"_blank\" rel=\"noopener ugc nofollow\">RFM (valor monet\u00e1rio de rec\u00eancia, frequ\u00eancia)<\/a>.<\/p>\n<p>Por isso, adicionamos um recurso <em>Rec\u00eancia desde a \u00faltima sess\u00e3o = 1 \/ N\u00famero de dias desde a \u00faltima sess\u00e3o <\/em>que permite que o valor seja normalizado entre 0 e 1<\/p>\n<\/div><div class=\"fusion-title title fusion-title-7 fusion-sep-none fusion-title-text fusion-title-size-three\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Recursos favoritos<\/h3><\/div><div class=\"fusion-text fusion-text-19\"><p>Tamb\u00e9m quer\u00edamos incluir algumas informa\u00e7\u00f5es sobre\u00a0<strong>a chave categ\u00f3rica data <\/strong>dispon\u00edveis, tais como <strong>navegador ou dispositivo<\/strong>. Como essas informa\u00e7\u00f5es est\u00e3o no n\u00edvel da sess\u00e3o, pode haver v\u00e1rios valores diferentes para um \u00fanico cliente, portanto, consideramos apenas aquele que ocorre mais por cliente (ou seja, o favorito). Al\u00e9m disso, para evitar ter recursos categ\u00f3ricos com cardinalidade muito alta, mantemos apenas os cinco valores mais comuns para cada recurso e substitu\u00edmos todos os outros valores por um valor \u201cOutro\u201d<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27589%27%20height%3D%270%27%20viewBox%3D%270%200%20589%200%27%3E%3Crect%20width%3D%27589%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article11.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 589px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"589\" height=\"auto\" \/><div class=\"fusion-title title fusion-title-8 fusion-sep-none fusion-title-text fusion-title-size-three\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Caracter\u00edsticas do produto<\/h3><\/div><div class=\"fusion-text fusion-text-20\"><p>Embora os dois primeiros tipos de recursos sejam definitivamente \u00fateis para nos ajudar a responder \u00e0 pergunta \u201cUm cliente vai comprar no meu site?\u201d, eles n\u00e3o s\u00e3o suficientemente espec\u00edficos se precisarmos saber<strong class=\"hq jr\">\u00a0\u201c<\/strong><strong>O cliente vai comprar um produto espec\u00edfico?<\/strong><strong class=\"hq jr\">\u201d<\/strong>. Para ajudar a responder a essa pergunta, criamos recursos espec\u00edficos do produto que incluem apenas o produto para o qual estamos tentando prever a compra:<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27623%27%20height%3D%270%27%20viewBox%3D%270%200%20623%200%27%3E%3Crect%20width%3D%27623%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article12.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 623px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"623\" height=\"auto\" \/><div class=\"fusion-text fusion-text-21\"><p>Para <em>Rec\u00eancia desde a \u00faltima sess\u00e3o com pelo menos uma intera\u00e7\u00e3o com esse produto,\u00a0<\/em>usamos a mesma f\u00f3rmula que para o\u00a0<em>Rec\u00eancia da sess\u00e3o<\/em>\u00a0nos Recursos gerais. No entanto, podemos ter casos em que h\u00e1 0 sess\u00e3o com pelo menos uma intera\u00e7\u00e3o com o produto e, nesse caso, preenchemos com 0. Isso faz sentido do ponto de vista comercial, pois o valor mais alto poss\u00edvel \u00e9 1 (quando o cliente teve uma sess\u00e3o desde ontem).<\/p>\n<\/div><div class=\"fusion-title title fusion-title-9 fusion-sep-none fusion-title-text fusion-title-size-three\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Caracter\u00edsticas de produtos similares<\/h3><\/div><div class=\"fusion-text fusion-text-22\"><p>Al\u00e9m de observar a intera\u00e7\u00e3o do cliente com o produto para o qual estamos tentando prever a probabilidade de compra, saber que o cliente interagiu com o<strong>\u00a0outros produtos com fun\u00e7\u00e3o e faixa de pre\u00e7o semelhantes\u00a0<\/strong>pode ser definitivamente \u00fatil<strong class=\"hq jr\">\u00a0<\/strong>(ou seja, produto substituto). Por esse motivo, adicionamos um conjunto de recursos de produtos similares que s\u00e3o id\u00eanticos aos recursos de produtos, exceto pelo fato de tamb\u00e9m incluirmos produtos similares no escopo da vari\u00e1vel. Os produtos similares de um determinado produto foram definidos usando inputs comerciais.<\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27608%27%20height%3D%270%27%20viewBox%3D%270%200%20608%200%27%3E%3Crect%20width%3D%27608%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article13.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 608px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"608\" height=\"auto\" \/><div class=\"fusion-text fusion-text-23\"><p>Agora temos nosso\u00a0<strong>dataset projetado com recursos<\/strong>\u00a0no qual podemos treinar nosso modelo de aprendizado de m\u00e1quina.<\/p>\n<\/div><div class=\"fusion-title title fusion-title-10 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">Treinamento do modelo<\/h2><\/div><div class=\"fusion-text fusion-text-24\"><p>Como queremos saber se um cliente vai comprar um produto espec\u00edfico ou n\u00e3o, esse \u00e9 um\u00a0<strong>problema de classifica\u00e7\u00e3o bin\u00e1ria.<\/strong><\/p>\n<p>Em nossa primeira itera\u00e7\u00e3o, fizemos o seguinte para criar nosso dataset de aprendizado de m\u00e1quina (que era uma linha por cliente):<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-4 fusion-checklist-default type-icons\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Calcule o <strong>recursos que utilizam as sess\u00f5es em um per\u00edodo de 3 meses<\/strong>\u00a0para cada cliente.<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Calcule o\u00a0<strong>meta de usar as sess\u00f5es em um per\u00edodo de 3 semanas<\/strong>\u00a0ap\u00f3s a janela de tempo do recurso. Se houver pelo menos uma compra do produto na janela de tempo, Target ser\u00e1 igual a 1 (definido como Classe 1), caso contr\u00e1rio, Target ser\u00e1 igual a 0 (definido como Classe 0)<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Divida o data entre um conjunto de treinamento e um teste usando a divis\u00e3o aleat\u00f3ria 80\/20.<\/div><\/li><\/ul><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27608%27%20height%3D%270%27%20viewBox%3D%270%200%20608%200%27%3E%3Crect%20width%3D%27608%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article14.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 608px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"608\" height=\"auto\" \/><div class=\"fusion-text fusion-text-25\"><p>No entanto, uma primeira explora\u00e7\u00e3o do data mostrou rapidamente que havia um<strong>\u00a0forte problema de desequil\u00edbrio de classe<\/strong>: A rela\u00e7\u00e3o Classe 1 \/ Classe 0 era superior a 1:1000 e n\u00e3o t\u00ednhamos clientes Classe 1 suficientes. Isso pode ser muito problem\u00e1tico para os modelos de aprendizado de m\u00e1quina.<\/p>\n<p>Para lidar com esses problemas, fizemos v\u00e1rias modifica\u00e7\u00f5es em nossa abordagem:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-5 fusion-checklist-default type-icons paddingList dark-text\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">N\u00f3s <strong>trocou a vari\u00e1vel de destino<\/strong> de fazer um\u00a0<strong>compra<\/strong>\u00a0para fazer um\u00a0<strong>adicionar ao carrinho<\/strong>. Portanto, nosso modelo perde um pouco em termos de significado comercial, mas o aumento do volume da Classe 1 mais do que compensa.<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">N\u00f3s <strong>treinou o modelo em v\u00e1rias janelas de deslocamento,<\/strong>cada uma de 3 meses + 3 semanas, em vez de uma \u00fanica. Al\u00e9m de aumentar nossos volumes de data, isso melhora a capacidade de generaliza\u00e7\u00e3o do modelo, treinando em v\u00e1rias \u00e9pocas do ano em que os clientes podem ter comportamentos de compra diferentes. Observe que, devido a isso, o mesmo cliente estar\u00e1 presente v\u00e1rias vezes no conjunto de data (em per\u00edodos diferentes). Para evitar vazamento de data, garantimos que ele esteja sempre no dataset de treinamento ou de teste.<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">N\u00f3s <strong>Subamostragem de nossa Classe 0, de modo que a propor\u00e7\u00e3o Classe 1 \/ Classe 0 seja 1<\/strong>. A subamostragem \u00e9 uma boa solu\u00e7\u00e3o para lidar com o problema de desequil\u00edbrio de classe, em compara\u00e7\u00e3o com outras op\u00e7\u00f5es, como a sobreamostragem ou a\u00a0<a class=\"bv kv\" href=\"https:\/\/machinelearningmastery.com\/smote-oversampling-for-imbalanced-classification\/\" target=\"_blank\" rel=\"noopener ugc nofollow\">SMOTE<\/a>, O senhor pode ter certeza de que o volume da Classe 1 foi aumentado consideravelmente com as duas primeiras altera\u00e7\u00f5es.\u00a0<strong>Somente o conjunto de treinamento \u00e9 rebalanceado<\/strong>j\u00e1 que queremos que o conjunto de teste tenha as mesmas propor\u00e7\u00f5es de classe do que o futuro data em que o testaremos. Observe que testamos com propor\u00e7\u00f5es mais altas, como 5 ou 10, mas 1 foi o ideal na avalia\u00e7\u00e3o do modelo.<\/div><\/li><\/ul><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27549%27%20height%3D%270%27%20viewBox%3D%270%200%20549%200%27%3E%3Crect%20width%3D%27549%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 549px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"549\" height=\"auto\" \/><div class=\"fusion-text fusion-text-26\"><p>Usando esse dataset, testamos v\u00e1rios modelos de classifica\u00e7\u00e3o: Modelo Linear, Random Forest e XGboost, ajustando os hiperpar\u00e2metros usando a busca em grade, e acabamos selecionando um<strong class=\"hq jr\">\u00a0<\/strong><strong><a class=\"bv kv\" href=\"https:\/\/xgboost.readthedocs.io\/en\/stable\/\" target=\"_blank\" rel=\"noopener ugc nofollow\">Modelo XGboost<\/a><\/strong><strong class=\"hq jr\">.<\/strong><\/p>\n<\/div><div class=\"fusion-title title fusion-title-11 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">Avalia\u00e7\u00e3o do nosso modelo<\/h2><\/div><div class=\"fusion-text fusion-text-27\"><p>Ao avaliar um modelo de propens\u00e3o, h\u00e1 dois tipos principais de avalia\u00e7\u00f5es que podem ser realizadas:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-6 fusion-checklist-default type-icons paddingList dark-text\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\"><strong>Avalia\u00e7\u00e3o de backtest<\/strong><\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">\n<p><strong>Avalia\u00e7\u00e3o de teste ao vivo<\/strong><\/p>\n<\/div><\/li><\/ul><div class=\"fusion-title title fusion-title-12 fusion-sep-none fusion-title-text fusion-title-size-three\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Avalia\u00e7\u00e3o de backtest<\/h3><\/div><div class=\"fusion-text fusion-text-28\"><p>Primeiro, realizamos\u00a0<strong>avalia\u00e7\u00e3o de backtest<\/strong>:<strong class=\"hq jr\">\u00a0<\/strong>aplicamos nosso modelo a <strong>hist\u00f3rico anterior data<\/strong>\u00a0e verificamos se o nosso modelo est\u00e1 identificando corretamente os clientes que v\u00e3o fazer uma adi\u00e7\u00e3o ao carrinho. Como estamos usando um classificador bin\u00e1rio, o modelo produz uma pontua\u00e7\u00e3o de probabilidade entre 0 e 1 de ser Classe 1 (Adicionar ao carrinho).<br \/>\n<a class=\"bv kv\" href=\"https:\/\/en.wikipedia.org\/wiki\/Confusion_matrix\" target=\"_blank\" rel=\"noopener ugc nofollow\">matriz de confus\u00e3o<\/a><strong class=\"hq jr\">\u00a0<\/strong>e calcular o\u00a0<strong><a class=\"bv kv\" href=\"https:\/\/en.wikipedia.org\/wiki\/Precision_and_recall\" target=\"_blank\" rel=\"noopener ugc nofollow\">precis\u00e3o\/recupera\u00e7\u00e3o<\/a>\u00a0<\/strong>(ou sua forma combinada no<strong><a class=\"bv kv\" href=\"https:\/\/en.wikipedia.org\/wiki\/F-score\" target=\"_blank\" rel=\"noopener ugc nofollow\">pontua\u00e7\u00e3o f1<\/a><\/strong>). No entanto, h\u00e1 dois problemas com essas m\u00e9tricas simples:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-7 fusion-checklist-default type-icons paddingList dark-text\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Alguns podem ser\u00a0<strong>dif\u00edcil de interpretar<\/strong>\u00a0porque o conjunto data est\u00e1 desequilibrado\u00a0<em>(por exemplo, a m\u00e9trica de precis\u00e3o geralmente ser\u00e1 muito baixa porque temos muito poucas classes 1)<\/em><\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Eles precisam decidir sobre um\u00a0<strong>limiar de probabilidade\u00a0<\/strong>para discriminar entre\u00a0<strong>Classes 0 e 1<\/strong><\/div><\/li><\/ul><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27575%27%20height%3D%270%27%20viewBox%3D%270%200%20575%200%27%3E%3Crect%20width%3D%27575%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article2-1.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 575px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"575\" height=\"auto\" \/><div class=\"fusion-text fusion-text-29\"><p>Por isso, decidimos usar duas m\u00e9tricas que eram mais\u00a0<strong>interpret\u00e1vel<\/strong>:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-8 fusion-checklist-default type-icons paddingList dark-text\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\"><strong>PR AUC: \u00c1rea sob a curva do gr\u00e1fico de precis\u00e3o por recupera\u00e7\u00e3o<\/strong>(<a href=\"https:\/\/neptune.ai\/blog\/f1-score-accuracy-roc-auc-pr-auc\" target=\"_blank\" rel=\"noopener ugc nofollow\">Veja esta explica\u00e7\u00e3o para obter mais detalhes<\/a>). Essencialmente, essa m\u00e9trica nos permite obter um <strong>avalia\u00e7\u00e3o global em todos os limites poss\u00edveis.<\/strong>Essa m\u00e9trica \u00e9 adequada para dataset desequilibrado, em que a prioridade \u00e9 maximizar a precis\u00e3o e a recupera\u00e7\u00e3o na classe minorit\u00e1ria: Classe 1 (ao contr\u00e1rio de sua prima, a ROC AUC)<\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\"><strong>Eleva\u00e7\u00e3o<\/strong>O senhor pode ver os resultados de sua pesquisa: classificamos os clientes por sua pontua\u00e7\u00e3o de probabilidade e dividimos nossos resultados em 20 ventila\u00e7\u00f5es. Uplift \u00e9 definido como o\u00a0<strong>Taxa da Classe 1 nos 5% principais \/ Taxa da Classe 1 em todos os dataset<\/strong>.\u00a0<em class=\"im\">Assim, por exemplo, se tivermos 21 % Add to Cart no Top 5 % do dataset contra 3 % Add to cart Rate em todo o dataset, teremos um aumento de 7, o que significa que nosso modelo \u00e9 7 vezes mais eficaz do que um modelo aleat\u00f3rio.<\/em><\/div><\/li><\/ul><div class=\"fusion-text fusion-text-30\"><p>Os resultados dessas m\u00e9tricas foram bastante positivos, principalmente,\u00a0<strong>Eleva\u00e7\u00e3o <\/strong>estava por perto\u00a0<strong>13.5.<\/strong><\/p>\n<p>A avalia\u00e7\u00e3o de backtest \u00e9 um m\u00e9todo livre de riscos para uma primeira avalia\u00e7\u00e3o de um modelo de propens\u00e3o, mas tem v\u00e1rias limita\u00e7\u00f5es:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-9 fusion-checklist-default type-icons paddingList dark-text\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Como isso \u00e9 feito apenas no passado, o resultado do modelo n\u00e3o est\u00e1 sendo realmente usado para <strong>impactar a estrat\u00e9gia de or\u00e7amento de m\u00eddia.<\/strong><\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">\n<p>Com nossas m\u00e9tricas, avaliamos apenas se o modelo era capaz de identificar corretamente os clientes que adicionariam ao carrinho, mas n\u00e3o avaliamos <strong>como a identifica\u00e7\u00e3o desses clientes geraria um aumento nas vendas.<\/strong><\/p>\n<\/div><\/li><\/ul><div class=\"fusion-title title fusion-title-13 fusion-sep-none fusion-title-text fusion-title-size-three\" style=\"--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">Avalia\u00e7\u00e3o de teste ao vivo<\/h3><\/div><div class=\"fusion-text fusion-text-31\"><p>Portanto, para ter uma ideia melhor do valor comercial do nosso modelo, precisamos realizar\u00a0<strong>avalia\u00e7\u00e3o de teste ao vivo. Aqui, ativamos nosso modelo e o usamos para priorizar os gastos com or\u00e7amento de publicidade:<\/strong><\/p>\n<\/div><img decoding=\"async\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27554%27%20height%3D%270%27%20viewBox%3D%270%200%20554%200%27%3E%3Crect%20width%3D%27554%27%20height%3D%270%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/12\/antoine-aubay-article3.png\" alt=\"Image\" class=\"lazyload artefact-elegant-image align-left hover-enable\" style=\"width: 554px; border-radius: 59% 41% 41% 59% \/ 29% 48% 52% 71%; overflow: hidden;\" width=\"554\" height=\"auto\" \/><div class=\"fusion-text fusion-text-32\"><p>Os resultados que obtivemos no teste ao vivo foram muito s\u00f3lidos:<\/p>\n<\/div><ul style=\"--awb-line-height:27.2px;--awb-icon-width:27.2px;--awb-icon-height:27.2px;--awb-icon-margin:11.2px;--awb-content-margin:38.4px;\" class=\"fusion-checklist fusion-checklist-10 fusion-checklist-default type-icons\"><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Em compara\u00e7\u00e3o com um\u00a0<strong>abordagem baseada em regras simples<\/strong> para propens\u00e3o de avalia\u00e7\u00e3o,\u00a0<strong>O ROAS de nosso modelo foi de +221 %<\/strong><\/div><\/li><li class=\"fusion-li-item\" style=\"\"><span class=\"icon-wrapper circle-no\"><i class=\"fusion-li-icon awb-icon-check\" aria-hidden=\"true\"><\/i><\/span><div class=\"fusion-li-item-content\">Al\u00e9m disso, tamb\u00e9m comparamos nosso desempenho com o de um forte concorrente na forma de<strong>\u00a0\u00cdndice de qualidade da sess\u00e3o do Google<\/strong>: uma pontua\u00e7\u00e3o fornecida pelo Google no Google Analytics dataset e, nesse caso\u00a0<strong>nosso modelo ainda estava em +73 % ROAS.<\/strong> Isso mostra como uma abordagem de ML personalizada pode trazer um valor comercial consider\u00e1vel.<\/div><\/li><\/ul><div class=\"fusion-title title fusion-title-14 fusion-sep-none fusion-title-text fusion-title-size-two\" style=\"--awb-margin-bottom-small:8px;\"><h2 class=\"fusion-title-heading title-heading-left fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:50;line-height:1.2;\">Conclus\u00e3o<\/h2><\/div><div class=\"fusion-text fusion-text-33\"><p>Al\u00e9m de alcan\u00e7ar um desempenho s\u00f3lido, um grande benef\u00edcio colateral de nossa abordagem \u00e9 que nossa engenharia de recursos \u00e9 muito gen\u00e9rica. Quase\u00a0<strong>Nenhuma das etapas de engenharia de recursos precisa ser adaptada<\/strong> para aplicar nosso modelo a um\u00a0<strong>escopo de pa\u00eds ou escopo de produto diferente<\/strong>. De fato, ap\u00f3s nosso primeiro sucesso no teste ao vivo, conseguimos <strong>implementar nosso modelo em v\u00e1rios pa\u00edses e produtos de maneira muito eficiente.<\/strong><\/p>\n<p>Obrigado ao senhor por ler. Ficaria feliz em ouvir os coment\u00e1rios do senhor sobre essa abordagem. O senhor j\u00e1 construiu modelos de propens\u00e3o? Se sim, o que o senhor fez de diferente?<\/p>\n<p><em>Agradecimentos a Bruce Delattre, Rafa\u00eblle Aygalenq e C\u00e9dric Ly.<\/em><\/p>\n<\/div><\/div><\/div><\/div><\/article><div class=\"fusion-fullwidth fullwidth-box fusion-builder-row-5 fusion-flex-container nonhundred-percent-fullwidth non-hundred-percent-height-scrolling\" style=\"--awb-border-radius-top-left:0px;--awb-border-radius-top-right:0px;--awb-border-radius-bottom-right:0px;--awb-border-radius-bottom-left:0px;--awb-margin-top:40px;--awb-margin-bottom:40px;--awb-flex-wrap:wrap;\" ><div class=\"fusion-builder-row fusion-row fusion-flex-align-items-center fusion-flex-justify-content-center fusion-flex-content-wrap\" style=\"max-width:calc( 1440px + 20px );margin-left: calc(-20px \/ 2 );margin-right: calc(-20px \/ 2 );\"><div class=\"fusion-layout-column fusion_builder_column fusion-builder-column-4 fusion_builder_column_1_1 1_1 fusion-flex-column fusion-flex-align-self-center\" style=\"--awb-padding-top:40px;--awb-padding-right:40px;--awb-padding-bottom:40px;--awb-padding-left:40px;--awb-overflow:hidden;--awb-bg-position:left center;--awb-bg-size:cover;--awb-border-color:rgba(10,17,40,0.1);--awb-border-style:solid;--awb-border-radius:4px 4px 4px 4px;--awb-width-large:100%;--awb-margin-top-large:0px;--awb-spacing-right-large:10px;--awb-margin-bottom-large:0px;--awb-spacing-left-large:10px;--awb-width-medium:100%;--awb-order-medium:0;--awb-spacing-right-medium:10px;--awb-spacing-left-medium:10px;--awb-width-small:100%;--awb-order-small:0;--awb-spacing-right-small:10px;--awb-spacing-left-small:10px;\"><div class=\"fusion-column-wrapper lazyload fusion-column-has-shadow fusion-flex-justify-content-center fusion-content-layout-column fusion-column-has-bg-image\" data-bg-url=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/03\/background.jpg\" data-bg=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/03\/background.jpg\"><div class=\"fusion-image-element\" style=\"text-align:center;--awb-margin-right:20px;--awb-margin-left:20px;--awb-max-width:150px;--awb-caption-title-font-family:var(--h2_typography-font-family);--awb-caption-title-font-weight:var(--h2_typography-font-weight);--awb-caption-title-font-style:var(--h2_typography-font-style);--awb-caption-title-size:var(--h2_typography-font-size);--awb-caption-title-transform:var(--h2_typography-text-transform);--awb-caption-title-line-height:var(--h2_typography-line-height);--awb-caption-title-letter-spacing:var(--h2_typography-letter-spacing);\"><span class=\"fusion-imageframe imageframe-none imageframe-2 hover-type-none\"><img decoding=\"async\" width=\"72\" height=\"41\" title=\"m\u00e9dio\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%2772%27%20height%3D%2741%27%20viewBox%3D%270%200%2072%2041%27%3E%3Crect%20width%3D%2772%27%20height%3D%2741%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/www.artefact.com\/\/wp-content\/uploads\/2021\/03\/medium.png\" alt class=\"lazyload img-responsive wp-image-60927\"\/><\/span><\/div><div class=\"fusion-title title fusion-title-15 fusion-sep-none fusion-title-center fusion-title-text fusion-title-size-three\" style=\"--awb-margin-top:20px;--awb-margin-bottom:0px;--awb-margin-bottom-small:8px;\"><h3 class=\"fusion-title-heading title-heading-center fusion-responsive-typography-calculated\" style=\"margin:0;--fontSize:20;line-height:1.2;\">M\u00e9dia Blog por Artefact.<\/h3><\/div><div class=\"fusion-text fusion-text-34\" style=\"--awb-content-alignment:center;\"><p>Este artigo foi publicado inicialmente no <strong>Medium.com<\/strong>.<br \/>\nSiga-nos em nosso Medium Blog !<\/p>\n<\/div><div style=\"text-align:center;\"><a class=\"fusion-button button-flat button-medium button-default fusion-button-default button-1 fusion-button-default-span fusion-button-default-type\" target=\"_blank\" rel=\"noopener noreferrer\" href=\"https:\/\/medium.com\/artefact-engineering-and-data-science\/scoring-customer-propensity-using-machine-learning-models-on-google-analytics-data-ba1126469c1f\"><span class=\"fusion-button-text awb-button__text awb-button__text--default\">Leia nosso artigo<\/span><\/a><\/div><\/div><\/div><\/div><\/div><\/p>","protected":false},"excerpt":{"rendered":"<p>Um mergulho profundo em como criamos modelos de aprendizado de m\u00e1quina personalizados de \u00faltima gera\u00e7\u00e3o para estimar a propens\u00e3o do cliente a comprar um produto usando o Google Analytics data.<\/p>","protected":false},"featured_media":68690,"parent":0,"template":"","meta":{"_acf_changed":false,"ep_exclude_from_search":false},"blog-category":[21939],"blog-language":[2991],"class_list":["post-65530","blog","type-blog","status-publish","has-post-thumbnail","hentry","blog-category-medium","blog-language-en"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog\/65530","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog"}],"about":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/types\/blog"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/media\/68690"}],"wp:attachment":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/media?parent=65530"}],"wp:term":[{"taxonomy":"blog-category","embeddable":true,"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog-category?post=65530"},{"taxonomy":"blog-language","embeddable":true,"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog-language?post=65530"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}