Abstract:Personalized dialogue systems have gained significant attention in recent years for their ability to generate responses in alignment with different personas. However, most existing approaches rely on pre-defined personal profiles, which are not only time-consuming and labor-intensive to create but also lack flexibility. We propose In-Dialogue Learning (IDL), a fine-tuning framework that enhances the ability of pre-trained large language models to leverage dialogue history to characterize persona for completing personalized dialogue generation tasks without pre-defined profiles. Our experiments on three datasets demonstrate that IDL brings substantial improvements, with BLEU and ROUGE scores increasing by up to 200% and 247%, respectively. Additionally, the results of human evaluations further validate the efficacy of our proposed method.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to generate conversations in line with the personality of a specific person in a personalized dialogue system without using predefined profiles. Most existing methods rely on predefined profiles to generate responses aligned with different personality traits. This method is not only time - consuming and labor - intensive, but also lacks flexibility. For this reason, the paper proposes the **In - Dialogue Learning (IDL)** framework, which enhances the ability of pre - trained large - language models to use dialogue history to characterize the personality of a person, thereby achieving the task of personalized dialogue generation without predefined profiles. Specifically, the paper mainly solves the following problems: 1. **Reducing dependence on predefined profiles**: By directly learning person information from dialogue sessions instead of relying on predefined profiles, the flexibility and adaptability of the system are improved. 2. **Improving the quality of personalized dialogue**: By proposing a two - stage method of **Mutual Supervised Learning (MSL)** and **Deep Personalized Alignment (DPA)**, the effect of personalized dialogue generation is optimized. Among them, MSL effectively organizes and utilizes personal information in the dialogue through static and dynamic person recognition technologies; DPA further optimizes the generation effect of personalized dialogue by introducing the **Direct Preference Optimization with Criterion (DPOC)** method and reduces the problem of preference degradation. 3. **Verifying the effectiveness of the method**: Through experiments on three datasets, the effectiveness of the IDL framework is proved, especially with a 200% increase in BLEU score and a 247% increase in ROUGE score respectively, and the human evaluation results further verify the effectiveness of the proposed method. In conclusion, the main contribution of this paper lies in providing a new method for generating high - quality personalized dialogue without predefined profiles, which significantly improves the performance and practicality of personalized dialogue systems.

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

Dialogue Learning with Human-in-the-Loop.

Learning through Dialogue Interactions by Asking Questions

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

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

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

Investigating deep reinforcement learning techniques in personalized dialogue generation

Personalizing a Dialogue System with Transfer Reinforcement Learning

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

I Know You Better: User Profile Aware Personalized Dialogue Generation

A Personalized Dialogue Generator with Implicit User Persona Detection

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

Exploring implicit persona knowledge for personalized dialogue generation

Personalized Dialogue Response Generation Learned from Monologues

Crafting a Good Prompt or Providing Exemplary Dialogues? A Study of In-Context Learning for Persona-based Dialogue Generation

Recent Trends in Personalized Dialogue Generation: A Review of Datasets, Methodologies, and Evaluations

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

Personalizing Task-oriented Dialog Systems via Zero-shot Generalizable Reward Function

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

A Pre-training Based Personalized Dialogue Generation Model with Persona-sparse Data

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