Improved Goal Oriented Dialogue via Utterance Generation and Look Ahead

TL;DR

This paper enhances goal-oriented dialogue systems by training a neural model to generate user utterances, improving intent prediction through multi-task learning and a novel look-ahead approach that disambiguates intents using generated utterances.

Contribution

It introduces a multi-task training regime utilizing unlabeled dialogue data and a look-ahead inference method with generated utterances to improve intent prediction accuracy.

Findings

Improved intent prediction accuracy over baseline models.

Effective use of unlabeled dialogue data for auxiliary training.

Enhanced disambiguation of user intents with generated utterances.

Abstract

Goal oriented dialogue systems have become a prominent customer-care interaction channel for most businesses. However, not all interactions are smooth, and customer intent misunderstanding is a major cause of dialogue failure. We show that intent prediction can be improved by training a deep text-to-text neural model to generate successive user utterances from unlabeled dialogue data. For that, we define a multi-task training regime that utilizes successive user-utterance generation to improve the intent prediction. Our approach achieves the reported improvement due to two complementary factors: First, it uses a large amount of unlabeled dialogue data for an auxiliary generation task. Second, it uses the generated user utterance as an additional signal for the intent prediction model. Lastly, we present a novel look-ahead approach that uses user utterance generation to improve intent prediction in inference time. Specifically, we generate counterfactual successive user utterances for conversations with ambiguous predicted intents, and disambiguate the prediction by reassessing the concatenated sequence of available and generated utterances.