Joint Learning of Interactive Spoken Content Retrieval and Trainable User Simulator

TL;DR

This paper introduces a jointly trained, learnable user simulator for interactive spoken content retrieval, eliminating the need for hand-crafted simulators and improving system performance and realism.

Contribution

It proposes a novel approach to train a user simulator jointly with the retrieval system, enhancing realism and effectiveness without relying on manually designed user models.

Findings

Learned user simulators outperform hand-crafted ones in reward metrics.

Simulated users behave more like real users, improving system training.

Joint training leads to better retrieval actions and outcomes.

Abstract

User-machine interaction is crucial for information retrieval, especially for spoken content retrieval, because spoken content is difficult to browse, and speech recognition has a high degree of uncertainty. In interactive retrieval, the machine takes different actions to interact with the user to obtain better retrieval results; here it is critical to select the most efficient action. In previous work, deep Q-learning techniques were proposed to train an interactive retrieval system but rely on a hand-crafted user simulator; building a reliable user simulator is difficult. In this paper, we further improve the interactive spoken content retrieval framework by proposing a learnable user simulator which is jointly trained with interactive retrieval system, making the hand-crafted user simulator unnecessary. The experimental results show that the learned simulated users not only achieve larger rewards than the hand-crafted ones but act more like real users.