# Multiple Generative Models Ensemble for Knowledge-Driven Proactive Human-Computer Dialogue Agent

## TL;DR

This paper presents an ensemble of multiple generative models for proactive human-computer dialogue, achieving significant improvements in response quality through data augmentation and novel ensemble techniques.

## Contribution

It introduces a rank-based ensemble approach combined with data augmentation and variant encoder-decoder structures for enhanced dialogue generation.

## Key findings

- Single model improves F1-score and BLEU by 18.67% over baseline
- Ensemble methods outperform baseline by 35.85%
- Effective for multi-turn proactive dialogue generation

## Abstract

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%.