A Systematic Evaluation of Response Selection for Open Domain Dialogue

TL;DR

This paper presents a new manually annotated dataset for response selection in open-domain dialogue, enabling a systematic evaluation of state-of-the-art methods and demonstrating significant performance improvements over synthetic data training.

Contribution

It introduces a manually annotated dataset for response ranking, aligning training data more closely with real-world use cases, and evaluates various response selection strategies.

Findings

Using multiple positive candidates improves response ranking.

Manually verified hard negatives enhance model performance.

Response selection models achieve up to 13% higher Recall@1.

Abstract

Recent progress on neural approaches for language processing has triggered a resurgence of interest on building intelligent open-domain chatbots. However, even the state-of-the-art neural chatbots cannot produce satisfying responses for every turn in a dialog. A practical solution is to generate multiple response candidates for the same context, and then perform response ranking/selection to determine which candidate is the best. Previous work in response selection typically trains response rankers using synthetic data that is formed from existing dialogs by using a ground truth response as the single appropriate response and constructing inappropriate responses via random selection or using adversarial methods. In this work, we curated a dataset where responses from multiple response generators produced for the same dialog context are manually annotated as appropriate (positive) and inappropriate (negative). We argue that such training data better matches the actual use case examples, enabling the models to learn to rank responses effectively. With this new dataset, we conduct a systematic evaluation of state-of-the-art methods for response selection, and demonstrate that both strategies of using multiple positive candidates and using manually verified hard negative candidates can bring in significant performance improvement in comparison to using the adversarial training data, e.g., increase of 3% and 13% in Recall@1 score, respectively.