| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 10.62 MB | Adobe PDF |
Authors
Abstract(s)
Com a constante evolução do mundo digital e tendo em consideração a situação pandémica atual, a utilização de emails como meio de comunicação apresenta um crescimento notório, assim como o email marketing, que é considerado o canal que gera o maior Return On Investment (ROI). Com as caixas de email sobrecarregadas, os utilizadores tendem a priorizar certos emails em detrimento de outros com base em algumas decisões, sendo a leitura do assunto um fator decisivo na sua abertura. Assim, pretende-se criar uma ferramenta de suporte aos editores das campanhas de email, que analisa e classifica um determinado assunto, de forma a aumentar a sua taxa de aberturas e consequente sucesso da campanha. Com base no histórico de emails enviados pela plataforma E-goi, procura-se criar um sistema que prevê o impacto de um determinado assunto, classificando-o de zero a cinco estrelas. Iniciado pela introdução ao projeto, na qual se identifica o problema existente e os objetivos propostos, e se apresenta a empresa na qual o projeto se enquadra, o presente documento visa contextualizar o leitor quanto aos conceitos mais relevantes relativos ao projeto e às soluções concorrentes existentes no mercado. De seguida, analisam-se técnicas de Natural Language Processing (NLP), assim como algoritmos de Machine Learning (ML) e Deep Learning (DL), de forma a concluir qual a melhor estratégia a adotar. Logo após, são apresentadas as frameworks de apoio à implementação mais utilizadas. Posteriormente, é elaborada a análise de valor de negócio que visa analisar e avaliar o propósito do projeto. Posto isto, é analisada e modelada a solução na qual se explica o domínio, os requisitos pretendidos e se apresentam duas soluções alternativas de design, seguidas da avaliação e escolha da mais adequada. Uma vez definida a melhor abordagem arquitetural, apresenta-se o design detalhado da solução, seguida da implementação da mesma. Por último, de modo a avaliar o potencial a nível funcional, a qualidade e usabilidade do sistema e a satisfação do utilizador, apresentam-se métodos de avaliação baseados em testes de software, testes de hipóteses, inquéritos e no modelo Quantitative Evaluation Framework (QEF).
Regarding the constant evolution of the digital world and considering the current pandemic situation, the emails usage as a communication channel presents notable growth, as well as email marketing, which is considered the channel that generates the highest ROI. With mailboxes overloaded, users tend to prioritize certain emails over others based on some decisions, being the reading of the subject a decisive factor in its opening. Thus, it is intended to create a support tool for the email campaigns editors, which analyzes and classifies a certain subject, in order to increase its openings rate and consequent success of the campaign. Based on the history of emails sent by the E-goi platform, an attempt is made to create a system that predicts the impact of a given subject, classifying it from zero to five stars. Initiated by the introduction of the project, which identifies the existing problem and the proposed objectives, and presents the company in which the project fits, this document aims to contextualize the reader regarding the most relevant concepts related to the project and the competing solutions existing in the market. Then, NLP techniques are analyzed as well as ML and DL algorithms in order to conclude which is the best strategy to adopt. Soon after, the most used implementation support frameworks are presented. Subsequently, the business value analysis is performed to analyze and evaluate the purpose of the project. Therefore, the solution is analyzed and modeled, explaining the domain and the requirements and presenting two alternative design solutions followed by the evaluation and choice of the most appropriate one. Once the best architectural approach is defined, the detailed design of the solution is presented, followed by its implementation. Finally, in order to assess the functional potential, quality and usability of the system and user satisfaction, evaluation methods based on software tests, hypothesis tests, surveys and the QEF model are presented.
Regarding the constant evolution of the digital world and considering the current pandemic situation, the emails usage as a communication channel presents notable growth, as well as email marketing, which is considered the channel that generates the highest ROI. With mailboxes overloaded, users tend to prioritize certain emails over others based on some decisions, being the reading of the subject a decisive factor in its opening. Thus, it is intended to create a support tool for the email campaigns editors, which analyzes and classifies a certain subject, in order to increase its openings rate and consequent success of the campaign. Based on the history of emails sent by the E-goi platform, an attempt is made to create a system that predicts the impact of a given subject, classifying it from zero to five stars. Initiated by the introduction of the project, which identifies the existing problem and the proposed objectives, and presents the company in which the project fits, this document aims to contextualize the reader regarding the most relevant concepts related to the project and the competing solutions existing in the market. Then, NLP techniques are analyzed as well as ML and DL algorithms in order to conclude which is the best strategy to adopt. Soon after, the most used implementation support frameworks are presented. Subsequently, the business value analysis is performed to analyze and evaluate the purpose of the project. Therefore, the solution is analyzed and modeled, explaining the domain and the requirements and presenting two alternative design solutions followed by the evaluation and choice of the most appropriate one. Once the best architectural approach is defined, the detailed design of the solution is presented, followed by its implementation. Finally, in order to assess the functional potential, quality and usability of the system and user satisfaction, evaluation methods based on software tests, hypothesis tests, surveys and the QEF model are presented.
Description
Keywords
Campanha de Email Assunto Classificação Taxa de Aberturas Natural Language Processing Machine Learning Deep Learning Email campaign Subject Classification Openings Rate Natural Language Processing Machine Learning Deep Learning