Publicação

Making open-ended questions attractive: leveraging topic modelling to evaluate survey responses

Ver documento

Detalhes bibliográficos
Resumo:While open-ended survey questions have declined due to time-intensive human analysis, ad vances in AI have renewed interest through the use of topic modeling. This thesis compares Latent Dirichlet Allocation, Non-negative Matrix Factorization, Top2Vec, BERTopic, and GPT-4.o in analyzing open-ended responses across two datasets (165, 170 answers). Model outputs are evaluated for theme replication of manual analyses and the level of human review required. Reviewed model outputs are also evaluated for theme replication supported by ex pert opinions. Results show that a human-in-the-loop approach with GPT-4.o requires mini mal manual review, closely replicates manual analysis, thus, effectively supports open-ended question analysis for organizations.
Autores principais:Schühle, Annika
Assunto:Natural language processing Topic modelling Thematic analysis GPT LDA NMF Top2Vec BERTopic Human-in-the-loop automated analysis
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso embargado
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL

Registos relacionados

Registos relacionados