| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.56 MB | Adobe PDF |
Advisor(s)
Abstract(s)
No último ano académico, os estudantes do ISEP necessitam de realizar um projeto final para
obtenção do grau académico que pretendem alcançar. O ISEP fornece uma plataforma digital
onde é possível visualizar todos os projetos que os alunos se podem candidatar. Apesar das
vantagens que a plataforma digital traz, esta também possui alguns problemas, nomeadamente
a difícil escolha de projetos adequados ao estudante devido à excessiva oferta e falta de
mecanismos de filtragem. Para além disso, existe também uma indecisão acrescida para
selecionar um supervisor que seja compatível para o projeto selecionado.
Tendo o aluno escolhido o projeto e o supervisor, dá-se início à fase de monitorização do
mesmo, que possui também os seus problemas, como o uso de diversas ferramentas que
posteriormente levam a possíveis problemas de comunicação e dificuldade em manter um
histórico de versões do trabalho desenvolvido.
De forma a responder aos problemas mencionados, realizou-se um estudo aprofundado dos
tópicos de sistemas de recomendação aplicados a Machine Learning e Learning Management
Systems. Para cada um desses grandes temas, foram analisados sistemas semelhantes capazes
de solucionar o problema proposto, tais como sistemas de recomendação desenvolvidos em
artigos científicos, aplicações comerciais e ferramentas como o ChatGPT.
Através da análise do estado da arte, concluiu-se que a solução para os problemas propostos
seria a criação de uma aplicação Web para alunos e supervisores, que juntasse as duas
temáticas analisadas. O sistema de recomendação desenvolvido possui filtragem colaborativa
com factorização de matrizes, e filtragem por conteúdo com semelhança de cossenos. As
tecnologias utilizadas no sistema centram-se em Python no back-end (com o uso de TensorFlow
e NumPy para funcionalidades de Machine Learning) e Svelte no front-end. O sistema foi
inspirado numa arquitetura em microsserviços em que cada serviço é representado pelo seu
próprio contentor de Docker, e disponibilizado ao público através de um domínio público.
O sistema foi avaliado através de três métricas: performance, confiabilidade e usabilidade. Foi
utilizada a ferramenta Quantitative Evaluation Framework para definir dimensões, fatores e
requisitos(e respetivas pontuações). Os estudantes que testaram a solução avaliaram o sistema
de recomendação com um valor de aproximadamente 7 numa escala de 1 a 10, e os valores de
precision, recall, false positive rate e F-Measure foram avaliados em 0.51, 0.71, 0.23 e 0.59
respetivamente. Adicionalmente, ambos os grupos classificaram a aplicação como intuitiva e
de fácil utilização, com resultados a rondar o 8 numa escala de 1 em 10.
In the last academic year, students at ISEP need to complete a final project to obtain the academic degree they aim to achieve. ISEP provides a digital platform where all the projects that students can apply for can be viewed. Besides the advantages this platform has, it also brings some problems, such as the difficult selection of projects suited for the student due to the excessive offering and lack of filtering mechanisms. Additionally, there is also increased difficulty in selecting a supervisor compatible with their project. Once the student has chosen the project and the supervisor, the monitoring phase begins, which also has its issues, such as using various tools that may lead to potential communication problems and difficulty in maintaining a version history of the work done. To address the mentioned problems, an in-depth study of recommendation systems applied to Machine Learning and Learning Management Systems was conducted. For each of these themes, similar systems that could solve the proposed problem were analysed, such as recommendation systems developed in scientific papers, commercial applications, and tools like ChatGPT. Through the analysis of the state of the art, it was concluded that the solution to the proposed problems would be the creation of a web application for students and supervisors that combines the two analysed themes. The developed recommendation system uses collaborative filtering with matrix factorization and content-based filtering with cosine similarity. The technologies used in the system are centred around Python on the backend (with the use of TensorFlow and NumPy for Machine Learning functionalities) and Svelte on the frontend. The system was inspired by a microservices architecture, where each service is represented by its own Docker container, and it was made available online through a public domain. The system was evaluated through performance, reliability, and usability. The Quantitative Evaluation Framework tool was used to define dimensions, factors, and requirements (and their respective scores). The students who tested the solution rated the recommendation system with a value of approximately 7 on a scale of 1 to 10, and the precision, recall, false positive rate, and F-Measure values were evaluated at 0.51, 0.71, 0.23, and 0.59, respectively. Additionally, both groups rated the application as intuitive and easy to use, with ratings around 8 on a scale of 1 to 10.
In the last academic year, students at ISEP need to complete a final project to obtain the academic degree they aim to achieve. ISEP provides a digital platform where all the projects that students can apply for can be viewed. Besides the advantages this platform has, it also brings some problems, such as the difficult selection of projects suited for the student due to the excessive offering and lack of filtering mechanisms. Additionally, there is also increased difficulty in selecting a supervisor compatible with their project. Once the student has chosen the project and the supervisor, the monitoring phase begins, which also has its issues, such as using various tools that may lead to potential communication problems and difficulty in maintaining a version history of the work done. To address the mentioned problems, an in-depth study of recommendation systems applied to Machine Learning and Learning Management Systems was conducted. For each of these themes, similar systems that could solve the proposed problem were analysed, such as recommendation systems developed in scientific papers, commercial applications, and tools like ChatGPT. Through the analysis of the state of the art, it was concluded that the solution to the proposed problems would be the creation of a web application for students and supervisors that combines the two analysed themes. The developed recommendation system uses collaborative filtering with matrix factorization and content-based filtering with cosine similarity. The technologies used in the system are centred around Python on the backend (with the use of TensorFlow and NumPy for Machine Learning functionalities) and Svelte on the frontend. The system was inspired by a microservices architecture, where each service is represented by its own Docker container, and it was made available online through a public domain. The system was evaluated through performance, reliability, and usability. The Quantitative Evaluation Framework tool was used to define dimensions, factors, and requirements (and their respective scores). The students who tested the solution rated the recommendation system with a value of approximately 7 on a scale of 1 to 10, and the precision, recall, false positive rate, and F-Measure values were evaluated at 0.51, 0.71, 0.23, and 0.59, respectively. Additionally, both groups rated the application as intuitive and easy to use, with ratings around 8 on a scale of 1 to 10.
Description
Keywords
Recommendation system Collaborative filtering Content-based filtering Project monitoring Student Supervisor