Publicação

Impact of cinematic aspects on ratings: A sentiment analysis of movie reviews

Ver documento

Detalhes bibliográficos
Resumo:Online movie reviews constitute a valuable source of audience feedback, yet their evaluative richness often exceeds the scope of overall sentiment analysis. Aspect-Based Sentiment Analysis (ABSA) provides a finer-grained approach by linking sentiments to specific cinematic dimensions. This dissertation investigates which aspects of films most strongly influence audience ratings, addressing the research question through a systematic, multi-phase methodology. A dataset of 1,000 IMDb reviews was used, with texts segmented and preprocessed before being analysed through three ABSA approaches. Two followed a pipeline structure, combining keyword-based aspect detection with sentiment classification via either the lexicon-based VADER model or the transformer-based DistilBERT, while the third adopted an end-to-end design using large language models (GPT-4o mini and GPT-4.1 mini) to extract aspects and sentiments directly. Performance was evaluated against a manually annotated gold standard for aspects and the dataset’s polarity labels for sentiment, using precision, recall, F1 score, and accuracy, with the comparative results guiding the selection of the most reliable method for subsequent analyses. With the selected method, review-level, dominant-aspect, and combination-aspect analyses were conducted. The findings demonstrate that Plot and Cast are most influential aspects in shaping ratings, while Directing and Ambience act as secondary but meaningful contributors. Moreover, results reveal that ratings emerge not from isolated dimensions but from the interaction of multiple aspects, particularly when Plot and Cast align. The study combines methodological rigor with applied analysis, providing academic insights into ABSA and offering practical implications for the film industry and streaming platforms.
Autores principais:Sousa, Catarina Filipa Lourenço de
Assunto:Sentiment analysis Aspect-based sentiment analysis (ABSA) Cinematic aspects Online reviews Movies Análise de sentimentos -- Sentiment analysis Análise de sentimentos baseada em aspetos Aspetos cinematográficos Comentários online Filmes -- Films
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:ISCTE
Idioma:inglês
Origem:Repositório ISCTE
Descrição
Resumo:Online movie reviews constitute a valuable source of audience feedback, yet their evaluative richness often exceeds the scope of overall sentiment analysis. Aspect-Based Sentiment Analysis (ABSA) provides a finer-grained approach by linking sentiments to specific cinematic dimensions. This dissertation investigates which aspects of films most strongly influence audience ratings, addressing the research question through a systematic, multi-phase methodology. A dataset of 1,000 IMDb reviews was used, with texts segmented and preprocessed before being analysed through three ABSA approaches. Two followed a pipeline structure, combining keyword-based aspect detection with sentiment classification via either the lexicon-based VADER model or the transformer-based DistilBERT, while the third adopted an end-to-end design using large language models (GPT-4o mini and GPT-4.1 mini) to extract aspects and sentiments directly. Performance was evaluated against a manually annotated gold standard for aspects and the dataset’s polarity labels for sentiment, using precision, recall, F1 score, and accuracy, with the comparative results guiding the selection of the most reliable method for subsequent analyses. With the selected method, review-level, dominant-aspect, and combination-aspect analyses were conducted. The findings demonstrate that Plot and Cast are most influential aspects in shaping ratings, while Directing and Ambience act as secondary but meaningful contributors. Moreover, results reveal that ratings emerge not from isolated dimensions but from the interaction of multiple aspects, particularly when Plot and Cast align. The study combines methodological rigor with applied analysis, providing academic insights into ABSA and offering practical implications for the film industry and streaming platforms.