Doing Personal LAPS: LLM-Augmented Dialogue Construction for Personalized Multi-Session Conversational Search

TL;DR

This paper tackles the scarcity of large-scale, multi-session, personalized dialogue data by introducing LAPS, a method that uses LLMs to guide a human worker in producing diverse, human-written conversations with explicit user preferences. It couples a semi-structured preference memory with an LLM-driven guidance and dialogue-act framework to collect, validate, and reuse user preferences across domains like recipes and movies. The approach yields a substantial, high-quality dataset with preference annotations, enabling improved personalized recommendations that leverage stored preferences rather than solely relying on historical dialogue. Results indicate that preference memory enhances explanation and relevance of recommendations and mitigates recall challenges in long-context prompts, while fully synthetic dialogue generation remains less diverse. The work offers a scalable path toward realistic personalized conversational systems and provides foundational data and methods for preference extraction and memory-based recommendation in multi-session settings.

Abstract

The future of conversational agents will provide users with personalized information responses. However, a significant challenge in developing models is the lack of large-scale dialogue datasets that span multiple sessions and reflect real-world user preferences. Previous approaches rely on experts in a wizard-of-oz setup that is difficult to scale, particularly for personalized tasks. Our method, LAPS, addresses this by using large language models (LLMs) to guide a single human worker in generating personalized dialogues. This method has proven to speed up the creation process and improve quality. LAPS can collect large-scale, human-written, multi-session, and multi-domain conversations, including extracting user preferences. When compared to existing datasets, LAPS-produced conversations are as natural and diverse as expert-created ones, which stays in contrast with fully synthetic methods. The collected dataset is suited to train preference extraction and personalized response generation. Our results show that responses generated explicitly using extracted preferences better match user's actual preferences, highlighting the value of using extracted preferences over simple dialogue history. Overall, LAPS introduces a new method to leverage LLMs to create realistic personalized conversational data more efficiently and effectively than previous methods.

Figures (5)

• Figure 1: A snippet from a multi-session dialogue. User preferences are extracted and stored in memory to generate personalized recommendations in subsequent sessions.
• Figure 2: Overview of our dialogue collection method (LAPS).
• Figure 3: Lexical Diversity of user and assistant utterances.
• Figure 4: LAPS collects diverse and high-quality dialogues compared to other dialogue collection methods.
• Figure 5: Breakdown of Preference Utilization ($\text{F}_{\text{PU}}$, recipe domain).