Classificando e-mails com serviços cognitivos

7 min readDec 12, 2021

Imagine ter acesso a uma plataforma na qual você pode desenvolver apps para seu negócios de forma rápida sem (ou pouca) codificação?
Isso já possível usando os serviços Azure, para entender mais sobre esse tipo de solução, leia mais aqui: Low-code application development on Azure

Em meu estudo de caso, criei um classificador de e-mails usando Azure Logic App, com a seguinte estrutura:

Descrição do fluxo:

Sempre que um novo email entra na caixa de entrada, um evento é disparado; Uma espécie de ETL rudimentar é iniciado no Logic App, gerando o fluxo de trabalho exibido acima. Em um fluxo perfeito, os emails passam por dois serviços cognitivos, o Text Analytics, e, o Content Moderator e por os dados catalogados por esses serviços são salvos em um diretório no Google Drive para serem usados transformados no dataset que disponibilizei no Kaggle: Mails Classification with Azure Cognitive

A grosso modo, o Logic App, nada mais é do que uma espécie de disparador de gatilhos em uma sequência lógica! //Meu Lego digital

Bom, após essa simplória e rápida introdução de como os dados foram gerados, vamos para nossa análise!

O que temos em nosso dataset ?

1.202 emails classificados.

E-mails recebidos entre 2020–04–21 à 2021–12–08

Segmentação por idioma:

Colunas quantitativas:

Text Analytics

Esse serviço cognitivo é usado basicamente para análise de texto bruto, suas funções são:

Análise de sentimentos Também conhecida como mineração de opinião, ou IA emocional é definida como o processo de análise de textos para determinar o tom emocional que é carregado, e foi classificado como positivo,neutro e negativo. //Esse foi o serviço que escolhi para classificação dos emails. 😎

Abaixo podemos ver que dos 1.2k emails classificados,51% recebeu um score de positivo. wow!

Mas e o mixed? 🤔
Será que isso se confirma na segmentação por idioma?

Agora iremos ver os scores definidos na mineração de opnião classificadas pelo serviço cognitivo segumentado por idioma. Esse score é chamado de pontuação de confiança, e quanto mais próximo de 1 for em cada uma das categorias "negativo", "neutro" e "positivo" uma média é gerada para a classificação na coluna sentiment.

🙄 Ok, mas e quando ele classifica um contéudo como mixed?

Olhando a documentação Cognitive Services - Text Analytics 3.0 Sentiment, a microsoft não deixa claro quando essa classificação hibrida ocorre. Mas ao meu ver é quando os score cai em um vale de indeterminância, o que pode contribuir para esse tipo de cenário é basicamente, um conteúdo com multiplos idiomas, excesso de erros gramaticais,e,ou ausência de conteudo.

E-mails classficiados como Mixed - Segmentado por idioma

unfold_moreShow hidden code

Núvem de palavras em inglês classificadas como possitivo.

Núvem de palavras em inglês classificadas como neutro.

Núvem de palavras em inglês classificadas como negativo.

Núvem de palavras em inglês classificadas como mixed

Analisando o resultado

From:		obfuscated@em.walmart.com
Subject:	🎄HOLI-YAY: Hot gifts under $25 this way 👉
Received:	2020-12-13T13:44:00.000000000---mail body---
Received:	Get presents FAST with free delivery on $35+ orders.‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌ My accountTechDealsToysDealsFashionDealsSportsDealsHomeDealsSeeAllPicks by interestTech  gadgets Tech  gadgets Toys Toys Food  cooking Food  cooking Unique gifts Unique gifts Find perfect presentsFor him For him For her For her For kids For kids For teens For teens For pets For pets For babies For babies Sleepwear for everyone. Women's Men's Kids'Sleepwear for everyone. Women's Men's Kids'  *Min. $35 order. Eligible items only.Please note that product availability is limited and prices are subject to change.Do not reply to this email. The inbox is not monitored. If you require assistance, pleaseclick here for help.UnsubscribeThis email is a Walmart advertisement. If you do not want to receive these email updates,click here to unsubscribe.Security  PrivacyWalmart protects your security and privacy. We will never ask for personal information (such as passwords or credit card numbers) in an email newsletter. If you receive such a request, please do not respond to the email.Learn more. See ourPrivacy Policy.Email Marketing, Walmart.com, 850 Cherry Avenue, San Bruno CA 94066©2020 Walmart.
---mail body---

Resultado da analise de sentimento

11 Sentenças identificadas:

positive 1
neutral 10
negative 1

Sentença 1 — Positiva

If you do not want to receive these email updates,click here to unsubscribe.Security PrivacyWalmart protects your security and privacy.

Confidence Scores:

positive: 0.82
neutral: 0.12
negative: 0.06

0.Get presents FAST with free delivery on35+orders.MyaccountTechDealsToysDealsFashionDealsSportsDealsHomeDealsSeeAllPicksbyinterestTechgadgetsTechgadgetsToysToysFoodcookingFoodcookingUniquegiftsUniquegiftsFindperfectpresentsForhimForhimForherForherForkidsForkidsForteensForteensForpetsForpetsForbabiesForbabiesSleepwearforeveryone.Women′sMen′sKids′Sleepwearforeveryone.Women′sMen′sKids′∗Min.35+orders.‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌MyaccountTechDealsToysDealsFashionDealsSportsDealsHomeDealsSeeAllPicksbyinterestTechgadgetsTechgadgetsToysToysFoodcookingFoodcookingUniquegiftsUniquegiftsFindperfectpresentsForhimForhimForherForherForkidsForkidsForteensForteensForpetsForpetsForbabiesForbabiesSleepwearforeveryone.Women′sMen′sKids′Sleepwearforeveryone.Women′sMen′sKids′∗Min.35 order. Eligible items only.Please note that product availability is limited and prices are subject to change.Do not reply to this email. The inbox is not monitored. If you require assistance, pleaseclick here for help.UnsubscribeThis email is a Walmart advertisement. If you do not want to receive these email updates,click here to unsubscribe.Security PrivacyWalmart protects your security and privacy. We will never ask for personal information (such as passwords or credit card numbers) in an email newsletter. If you receive such a request, please do not respond to the email.Learn more. See ourPrivacy Policy.Email Marketing, Walmart.com, 850 Cherry Avenue, San Bruno CA 94066©2020 Walmart.

Sentença 2 — Neutra

Eligible items only.Please note that product availability is limited and prices are subject to change.Do not reply to this email

Confidence Scores:

positive: 0.02
neutral: 0.65
negative: 0.33

0.Get presents FAST with free delivery on35+orders.MyaccountTechDealsToysDealsFashionDealsSportsDealsHomeDealsSeeAllPicksbyinterestTechgadgetsTechgadgetsToysToysFoodcookingFoodcookingUniquegiftsUniquegiftsFindperfectpresentsForhimForhimForherForherForkidsForkidsForteensForteensForpetsForpetsForbabiesForbabiesSleepwearforeveryone.Women′sMen′sKids′Sleepwearforeveryone.Women′sMen′sKids′∗Min.35+orders.‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌MyaccountTechDealsToysDealsFashionDealsSportsDealsHomeDealsSeeAllPicksbyinterestTechgadgetsTechgadgetsToysToysFoodcookingFoodcookingUniquegiftsUniquegiftsFindperfectpresentsForhimForhimForherForherForkidsForkidsForteensForteensForpetsForpetsForbabiesForbabiesSleepwearforeveryone.Women′sMen′sKids′Sleepwearforeveryone.Women′sMen′sKids′∗Min.35 order. Eligible items only.Please note that product availability is limited and prices are subject to change.Do not reply to this email. The inbox is not monitored. If you require assistance, pleaseclick here for help.UnsubscribeThis email is a Walmart advertisement. If you do not want to receive these email updates,click here to unsubscribe.Security PrivacyWalmart protects your security and privacy. We will never ask for personal information (such as passwords or credit card numbers) in an email newsletter. If you receive such a request, please do not respond to the email.Learn more. See ourPrivacy Policy.Email Marketing, Walmart.com, 850 Cherry Avenue, San Bruno CA 94066©2020 Walmart.

Outros sentenças encontradas

Neutral
Confidence Scores:
Positive: 0.06
Neutral: 0.93
Negavite: 0.01
Get presents FAST with free delivery on $35+ orders.‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌‌ My accountTechDealsToysDealsFashionDealsSportsDealsHomeDealsSeeAllPicks by interestTech gadgets Tech gadgets Toys Toys Food cooking Food cooking Unique gifts Unique gifts Find perfect presentsFor him For him For her For her For kids For kids For teens For teens For pets For pets For babies For babies Sleepwear for everyone.

Neutral

Confidence Scores:
Positive: 0.04
Neutral: 0.92
Negavite: 0.04
Women’s Men’s Kids’Sleepwear for everyone.

Neutral

Confidence Scores:
Positive: 0
Neutral: 0.99
Negavite: 0.01
Women’s Men’s Kids’ *Min.

Neutral

Confidence Scores:
Positive: 0.02
Neutral: 0.97
Negavite: 0.01
35 order

Neutral

Confidence Scores:
Positive: 0.02
Neutral: 0.65
Negavite: 0.33
Eligible items only.Please note that product availability is limited and prices are subject to change.Do not reply to this email.

Neutral

Confidence Scores:
Positive: 0.01
Neutral: 0.74
Negavite: 0.25
The inbox is not monitored.

Neutral

Confidence Scores:
Positive: 0.08
Neutral: 0.88
Negavite: 0.04
If you require assistance, pleaseclick here for help.UnsubscribeThis email is a Walmart advertisement.

Neutral

Confidence Scores:
Positive: 0.82
Neutral: 0.12
Negavite: 0.06
If you do not want to receive these email updates,click here to unsubscribe.Security PrivacyWalmart protects your security and privacy.

Neutral

Confidence Scores:
Positive: 0.01
Neutral: 0.96
Negavite: 0.03
We will never ask for personal information (such as passwords or credit card numbers) in an email newsletter.

Neutral