Publications
カンファレンス (国際) Movie Keyword Search Using Large-Scale Language Model With User-Generated Rankings and Reviews
Tensho Miyashita (Aoyama Gakuin univ.), Yoshiyuki Shoji (Aoyama Gakuin univ.), Sumio Fujita, Martin J. Dürst (Aoyama Gakuin univ.)
The 25th International Conference on Information Integration and Web Intelligence (iiWAS2023)
2023.11.22
This paper proposes a method for movie keyword searches using consumer-generated rankings and reviews. This method uses bidirectional encoder representations from transformers (BERT), a pretrained large-scale language model with task-oriented fine-tuning. The model was trained using paired ranked movie titles and their corresponding reviews. This fine-tuning enabled the model to estimate whether a given movie is likely to appear in a ranking that includes a given keyword. To complete this binary classification task, the method ranked the movies using probability. Using data from an accurate, well-known Japanese movie review site, we tuned the model using 10,000 user-generated rankings and 15,000 movies for an experiment. The experimental results showed that the proposed method achieved greater accuracy than existing similaritybased methods, although there is room for improvement in the pooling methods and other aspects.
Paper :
Movie Keyword Search Using Large-Scale Language Model With User-Generated Rankings and Reviews
（外部サイト）