Personalized LLM for Generating Customized Responses to the Same Query from Different Users

TL;DR

This paper introduces a novel querier-aware personalization method for large language models, generating distinct responses for the same query from different users by leveraging a dual-tower architecture and contrastive learning, supported by a new multilingual dataset.

Contribution

It proposes a new querier-aware personalization approach with a dual-tower model and contrastive learning, and creates a multi-querier dataset for training and evaluation.

Findings

Significant improvement in response quality, with ROUGE-L scores up to 48.7%.

High success rates in personalized response generation, ranging from 54% to 82%.

Effective clustering reduces query diversity impact on learning.

Abstract

Existing work on large language model (LLM) personalization assigned different responding roles to LLMs, but overlooked the diversity of queriers. In this work, we propose a new form of querier-aware LLM personalization, generating different responses even for the same query from different queriers. We design a dual-tower model architecture with a cross-querier general encoder and a querier-specific encoder. We further apply contrastive learning with multi-view augmentation, pulling close the dialogue representations of the same querier, while pulling apart those of different queriers. To mitigate the impact of query diversity on querier-contrastive learning, we cluster the dialogues based on query similarity and restrict the scope of contrastive learning within each cluster. To address the lack of datasets designed for querier-aware personalization, we also build a multi-querier dataset from English and Chinese scripts, as well as WeChat records, called MQDialog, containing 173 queriers and 12 responders. Extensive evaluations demonstrate that our design significantly improves the quality of personalized response generation, achieving relative improvement of 8.4% to 48.7% in ROUGE-L scores and winning rates ranging from 54% to 82% compared with various baseline methods.