Publication
Exploring Effective Information Utilization in Multi-Turn Topic-Driven Conversations
Combining topics and dialogue history through Fusion-in-Decoder for multi-turn conversation.
Cite this work View paper
Citation
Li, J., He, B., & Mi, F. (2022). Exploring Effective Information Utilization in Multi-Turn Topic-Driven Conversations. arXiv preprint arXiv:2209.00250.
About this work
This work uses Fusion-in-Decoder to combine topic information with dialogue history for multi-turn conversation. It compares three channel settings for encoding these inputs.
Experiments on the Chinese NaturalConv dataset examine news-grounded conversations. Combining a whole-passage channel with an additional history channel gives competitive results while accommodating long input texts.