Generating Informative and Diverse Conversational Responses via Adversarial Information Maximization

TL;DR

This paper introduces AIM, an adversarial learning approach that enhances both informativeness and diversity in neural conversational responses by matching response distributions and maximizing mutual information.

Contribution

The paper proposes a novel adversarial training framework that explicitly optimizes mutual information to improve response informativeness and uses distributional matching for diversity.

Findings

Significantly improves response informativeness and diversity.

Outperforms baseline models in automatic and human evaluations.

Effectively balances informativeness and diversity in conversational responses.

Abstract

Responses generated by neural conversational models tend to lack informativeness and diversity. We present Adversarial Information Maximization (AIM), an adversarial learning strategy that addresses these two related but distinct problems. To foster response diversity, we leverage adversarial training that allows distributional matching of synthetic and real responses. To improve informativeness, our framework explicitly optimizes a variational lower bound on pairwise mutual information between query and response. Empirical results from automatic and human evaluations demonstrate that our methods significantly boost informativeness and diversity.