Publicação
Temporal modelling for movie success prediction: a comparative study on the applicability of different deep learning models in entertainment analytics
| Resumo: | This thesis investigates factors influencing movie success using deep learning techniques. It compares weak supervision to supervised deep learning approaches when predicting movie success after the box office release. A key focus of this study is to establish whether the multiple instance learning (MIL) approach can perform more accurate predictions using multivariate time series data from the post-release stage of movies. Additionally, it is assessed whether a MIL-based architecture delivers more interpretable results compared to supervised architectures. By integrating diverse data sources and features, this study provides a comprehensive perspective on how different deep learning techniques can be used to measure audience engagement as a time series to classify box office success. The findings advance academic understanding of the applicability of MIL in the movie domain and offer insights for industry stakeholders aiming to enhance deep learning architecture selection. |
|---|---|
| Autores principais: | Pellinger, Luc Marcel |
| Assunto: | Multivariate Time Series Classification Movie success prediction Model comparison Multiple instance learning |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This thesis investigates factors influencing movie success using deep learning techniques. It compares weak supervision to supervised deep learning approaches when predicting movie success after the box office release. A key focus of this study is to establish whether the multiple instance learning (MIL) approach can perform more accurate predictions using multivariate time series data from the post-release stage of movies. Additionally, it is assessed whether a MIL-based architecture delivers more interpretable results compared to supervised architectures. By integrating diverse data sources and features, this study provides a comprehensive perspective on how different deep learning techniques can be used to measure audience engagement as a time series to classify box office success. The findings advance academic understanding of the applicability of MIL in the movie domain and offer insights for industry stakeholders aiming to enhance deep learning architecture selection. |
|---|