@inproceedings{zhu-etal-2023-towards-optimizing,
    title = "Towards Optimizing Pre-trained Language Model Ensemble Learning for Task-oriented Dialogue System",
    author = "Zhu, Zhiyuan  and
      Liao, Yusheng  and
      Chen, Zhe  and
      Wang, Yu  and
      Guan, Yunfeng",
    editor = "Chen, Yun-Nung  and
      Crook, Paul  and
      Galley, Michel  and
      Ghazarian, Sarik  and
      Gunasekara, Chulaka  and
      Gupta, Raghav  and
      Hedayatnia, Behnam  and
      Kottur, Satwik  and
      Moon, Seungwhan  and
      Zhang, Chen",
    booktitle = "Proceedings of the Eleventh Dialog System Technology Challenge",
    month = sep,
    year = "2023",
    address = "Prague, Czech Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/fix-sig-urls/2023.dstc-1.17/",
    pages = "144--149",
    abstract = "Task-oriented dialogue systems that employ external knowledge to generate informative responses have become an important field of research. This paper outlines our contribution to Track 5 of the Eleventh Dialog System Technology Challenge (DSTC11), which focuses on constructing high-performing, subjective knowledge-enriched task-oriented dialogue systems. Specifically, we investigate the complementarity of various language models to tackle the diverse knowledge selection task that involves multiple external sources. Based on this investigation, we propose pre- and post-generation model ensemble approaches to mitigate potential biases inherent in using a single model for the knowledge selection task. Finally, we utilize the consensus decoding approach to combine fine-tuned ensemble models and improve the performance of the generation system. Our system ranked 1st in human evaluation, even outperforming human annotation."
}