Relevance in Dialogue: Is Less More? An Empirical Comparison of Existing Metrics, and a Novel Simple Metric

TL;DR

This paper evaluates existing dialogue relevance metrics, finds their dependency on datasets, and introduces a simple, novel metric that improves correlation with human judgments while reducing dataset sensitivity without fine-tuning models.

Contribution

The authors propose a new simple relevance metric that outperforms existing metrics on multiple datasets, with less data and domain sensitivity, without requiring model fine-tuning.

Findings

Proposed metric achieves state-of-the-art on HUMOD dataset.

Reduces dataset sensitivity by 37%-66%.

Demonstrates competitive performance across four diverse datasets.

Abstract

In this work, we evaluate various existing dialogue relevance metrics, find strong dependency on the dataset, often with poor correlation with human scores of relevance, and propose modifications to reduce data requirements and domain sensitivity while improving correlation. Our proposed metric achieves state-of-the-art performance on the HUMOD dataset while reducing measured sensitivity to dataset by 37%-66%. We achieve this without fine-tuning a pretrained language model, and using only 3,750 unannotated human dialogues and a single negative example. Despite these limitations, we demonstrate competitive performance on four datasets from different domains. Our code, including our metric and experiments, is open sourced.