Reproduce “Topic-switch adapted Japanese Dialogue System based on PLATO-2” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Donghuo Zeng, Jianming Wu, Yanan Wang, Kazunori Matsumoto, Gen Hattori, Kazushi Ikeda
Researchers at KDDI Research, Inc. developed PLATO-JDS, a large-scale open-domain Japanese dialogue system built upon the PLATO-2 architecture, alongside a substantial Japanese dialogue dataset (Dialogue-Graph) of 1.656 million data points. The system, enhanced by a topic-switch algorithm, achieved an average human evaluation score of 1.767 out of 2.000 in human-bot chats, improving dialogue coherence, informativeness, engagingness, and humanness compared to the baseline PLATO-JDS.
Key insights
- The PLATO-2 architecture is effectively adaptable for building high-quality, open-domain dialogue systems in Japanese, demonstrating its cross-lingual potential.
- The creation of a comprehensive, large-scale Japanese dialogue dataset is fundamental for advancing the performance of conversational AI in this language.
- A dedicated topic-switch mechanism, driven by a BERT-based discriminator, is essential for improving dialogue coherence and user satisfaction by allowing the system to intelligently manage context during topic changes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.