Abstract:A task-oriented dialogue system (TOD) is an important application of artificial intelligence. In the past few years, works on personalized TODs have attracted increased research attention and have seen much progress. The main challenge of such dialogue systems is finding ways to exploit user profiles under conditions of fixed and monolithic dialogue style. However, most of the existing works overlook the observation that the personalization capability of the dialogue system is fundamental and generic, and they treat all attributes of the user profile equally throughout the dialogue flows, which makes them inadequate for developing a well-performing personalized TOD. In this paper, we propose a two-stage learning framework equipped with GPT2 as a backbone to alleviate the above two problems. In the first learning phase, we finetune the GPT2 model on the personalized open-domain dialogues to preliminarily acquire personalization power. Then, we transfer this power to personalized task-oriented dialogues for the second stage of learning. After that, we enable the proposed model to be capable of the desired personalization capacity. Moreover, we present a dynamic profile fusion mechanism and an auxiliary task that detects which attributes contribute to the current utterance to facilitate the model's performance. Eventually, we rewrite the attribute descriptions of user profiles in sentences to mitigate the consistency gap between the open-domain and task-oriented dialogues. The experimental results show that the proposed model achieves superior results compared to the state-of-the-art models on two versions of the Personalized bAbI dataset.

I Know You Better: User Profile Aware Personalized Dialogue Generation

Personalized Dialogue Response Generation Learned from Monologues

Persona-aware Dialogue Generation with Enriched Profile

Learning Personalized End-to-End Goal-Oriented Dialog.

User Profile Gender : Female , Age : Elderly , Dietary : Vegetable Good morning

A Personalized Dialogue Generator with Implicit User Persona Detection

Dialogue History Matters! Personalized Response Selection in Multi-Turn Retrieval-Based Chatbots

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

Dialogue History Matters! Personalized Response Selectionin Multi-turn Retrieval-based Chatbots

One Chatbot Per Person: Creating Personalized Chatbots based on Implicit User Profiles

Generating Personalized Dialogue via Multi-Task Meta-Learning

Apollonion: Profile-centric Dialog Agent

Diverse dialogue generation by fusing mutual persona-aware and self-transferrer

Bilateral personalized dialogue generation with contrastive learning

Learning Implicit User Profiles for Personalized Retrieval-Based Chatbot

"In Dialogues We Learn": Towards Personalized Dialogue Without Pre-defined Profiles through In-Dialogue Learning

Learning to Know Myself: A Coarse-to-Fine Persona-Aware Training Framework for Personalized Dialogue Generation

Multitask Learning and Reinforcement Learning for Personalized Dialog Generation: an Empirical Study.

Content-Oriented User Modeling for Personalized Response Ranking in Chatbots

Exploring implicit persona knowledge for personalized dialogue generation

From easy to hard: Improving personalized response generation of task-oriented dialogue systems by leveraging capacity in open-domain dialogues