The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems

TL;DR

The paper presents the Ubuntu Dialogue Corpus, a large-scale dataset of nearly 1 million multi-turn dialogues designed to advance research in neural dialogue systems, along with baseline neural architectures and benchmark results.

Contribution

It introduces a substantial, unstructured multi-turn dialogue dataset and provides neural models and benchmarks for response selection tasks.

Findings

The dataset contains over 7 million utterances and 100 million words.

Baseline neural models achieve measurable performance on response selection.

The dataset combines properties of structured dialogue and unstructured social media interactions.

Abstract

This paper introduces the Ubuntu Dialogue Corpus, a dataset containing almost 1 million multi-turn dialogues, with a total of over 7 million utterances and 100 million words. This provides a unique resource for research into building dialogue managers based on neural language models that can make use of large amounts of unlabeled data. The dataset has both the multi-turn property of conversations in the Dialog State Tracking Challenge datasets, and the unstructured nature of interactions from microblog services such as Twitter. We also describe two neural learning architectures suitable for analyzing this dataset, and provide benchmark performance on the task of selecting the best next response.