BERT-Based Movie Keyword Search Leveraging User-Generated Movie Rankings and Reviews

This paper introduces a novel method for movie keyword searches based on user-generated rankings and reviews. We utilize the capabilities of the BERT language model, which has been enriched with task-specific fine-tuning. The model is trained to understand the relationship between keywords and movies using paired user-generated ranking titles and movie reviews. We sourced our data from a renowned Japanese movie review platform. This dataset comprises 10,000 user rankings and 15,000 films. In a binary classification task, our model demonstrated superior performance compared to traditional similarity-based methods. While our approach outperforms traditional similarity methods, further improvements in pooling techniques are necessary.