Multiple Generative Models Ensemble for Knowledge-Driven Proactive Human-Computer Dialogue Agent
This work addresses dialogue agents for human-computer interaction, but it is incremental as it builds on existing models with ensemble techniques.
The paper tackled proactive dialogue generation by developing an ensemble of multiple sequence-to-sequence models, achieving an 18.67% average improvement in F1-score and BLEU over the baseline on the DuConv dataset, with ensemble methods boosting this to 35.85%.
Multiple sequence to sequence models were used to establish an end-to-end multi-turns proactive dialogue generation agent, with the aid of data augmentation techniques and variant encoder-decoder structure designs. A rank-based ensemble approach was developed for boosting performance. Results indicate that our single model, in average, makes an obvious improvement in the terms of F1-score and BLEU over the baseline by 18.67% on the DuConv dataset. In particular, the ensemble methods further significantly outperform the baseline by 35.85%.