Modeling Psychotherapy Dialogues with Kernelized Hashcode Representations: A Nonparametric Information-Theoretic Approach

TL;DR

This paper introduces a nonparametric kernelized hashcode framework for dialogue modeling that efficiently handles small to large datasets, improves mutual information-based model selection, and outperforms neural network models in psychotherapy dialogue tasks.

Contribution

It presents the first nonparametric kernel-based dialogue modeling approach using hashcodes, with a novel mutual information lower bound for better response prediction and alignment.

Findings

Outperforms neural network dialogue systems in response quality.

Reduces training time from days to hours.

Achieves higher human-evaluated response quality.

Abstract

We propose a novel dialogue modeling framework, the first-ever nonparametric kernel functions based approach for dialogue modeling, which learns kernelized hashcodes as compressed text representations; unlike traditional deep learning models, it handles well relatively small datasets, while also scaling to large ones. We also derive a novel lower bound on mutual information, used as a model-selection criterion favoring representations with better alignment between the utterances of participants in a collaborative dialogue setting, as well as higher predictability of the generated responses. As demonstrated on three real-life datasets, including prominently psychotherapy sessions, the proposed approach significantly outperforms several state-of-art neural network based dialogue systems, both in terms of computational efficiency, reducing training time from days or weeks to hours, and the response quality, achieving an order of magnitude improvement over competitors in frequency of being chosen as the best model by human evaluators.