Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Less is More: Learning to Refine Dialogue History for Personalized Dialogue Generation

Hanxun Zhong, Zhicheng Dou, Yutao Zhu, Hongjin Qian, Ji-Rong Wen

TL;DR

This work tackles the challenge of personalized dialogue generation from long, noisy dialogue histories. It proposes MSP, a hierarchical refiner framework with a user refiner, topic refiner, and token refiner feeding a transformer-based generator, guided by a joint training scheme and a supplementary sentence-matching task. Experiments on Weibo and Reddit demonstrate MSP's superior performance across metric-based and human evaluations, with ablations confirming the necessity of each refinement step and the benefit of cross-user information. The approach enables more informative, personalized responses by leveraging broader context while controlling noise, offering practical gains for large-scale personalized dialogue systems.

Abstract

Personalized dialogue systems explore the problem of generating responses that are consistent with the user's personality, which has raised much attention in recent years. Existing personalized dialogue systems have tried to extract user profiles from dialogue history to guide personalized response generation. Since the dialogue history is usually long and noisy, most existing methods truncate the dialogue history to model the user's personality. Such methods can generate some personalized responses, but a large part of dialogue history is wasted, leading to sub-optimal performance of personalized response generation. In this work, we propose to refine the user dialogue history on a large scale, based on which we can handle more dialogue history and obtain more abundant and accurate persona information. Specifically, we design an MSP model which consists of three personal information refiners and a personalized response generator. With these multi-level refiners, we can sparsely extract the most valuable information (tokens) from the dialogue history and leverage other similar users' data to enhance personalization. Experimental results on two real-world datasets demonstrate the superiority of our model in generating more informative and personalized responses.

Less is More: Learning to Refine Dialogue History for Personalized Dialogue Generation

TL;DR

This work tackles the challenge of personalized dialogue generation from long, noisy dialogue histories. It proposes MSP, a hierarchical refiner framework with a user refiner, topic refiner, and token refiner feeding a transformer-based generator, guided by a joint training scheme and a supplementary sentence-matching task. Experiments on Weibo and Reddit demonstrate MSP's superior performance across metric-based and human evaluations, with ablations confirming the necessity of each refinement step and the benefit of cross-user information. The approach enables more informative, personalized responses by leveraging broader context while controlling noise, offering practical gains for large-scale personalized dialogue systems.

Abstract

Personalized dialogue systems explore the problem of generating responses that are consistent with the user's personality, which has raised much attention in recent years. Existing personalized dialogue systems have tried to extract user profiles from dialogue history to guide personalized response generation. Since the dialogue history is usually long and noisy, most existing methods truncate the dialogue history to model the user's personality. Such methods can generate some personalized responses, but a large part of dialogue history is wasted, leading to sub-optimal performance of personalized response generation. In this work, we propose to refine the user dialogue history on a large scale, based on which we can handle more dialogue history and obtain more abundant and accurate persona information. Specifically, we design an MSP model which consists of three personal information refiners and a personalized response generator. With these multi-level refiners, we can sparsely extract the most valuable information (tokens) from the dialogue history and leverage other similar users' data to enhance personalization. Experimental results on two real-world datasets demonstrate the superiority of our model in generating more informative and personalized responses.
Paper Structure (34 sections, 12 equations, 3 figures, 7 tables, 1 algorithm)

This paper contains 34 sections, 12 equations, 3 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Personalized Dialogue Generation
  4. Retrieval-guided Natural Language Generation
  5. Methodology
  6. Problem Statement and Overview
  7. User Refiner
  8. Topic Refiner
  9. Token Refiner
  10. Generator
  11. Training and Optimization
  12. Supplementary Sentence Matching Task
  13. Joint Training
  14. Experiments
  15. Datasets
  16. ...and 19 more sections

Figures (3)

  • Figure 1: The overview structure of the proposed model which consists of four modules: (1) user refiner, (2) topic refiner, (3) token refiner, and (4) generator.
  • Figure 2: Comparison with the retrieval-based model on the Weibo dataset.
  • Figure 3: Experiments with the different number of user profiles on the Weibo dataset. To keep the dimension consistent, P-Cover is multiplied by a factor of 100.