Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Personalize Before Retrieve: LLM-based Personalized Query Expansion for User-Centric Retrieval

Yingyi Zhang, Pengyue Jia, Derong Xu, Yi Wen, Xianneng Li, Yichao Wang, Wenlin Zhang, Xiaopeng Li, Weinan Gan, Huifeng Guo, Yong Liu, Xiangyu Zhao

TL;DR

The paper tackles the challenge of personalized query expansion in retrieval-augmented generation by proposing PBR, a framework that pre-emptively adapts queries using user-specific signals. It combines P-PRF, which generates style-aligned pseudo utterances and reasoning, with P-Anchor, which grounds queries in a semantic graph of the user’s private corpus via Personalized PageRank. The fusion step dynamically weights these signals to produce a personalized query representation, leading to improved retrieval performance on PersonaBench and LongMemEval across multiple retrievers. Experiments and ablations demonstrate that pre-retrieval personalization yields substantial gains in both recall and ranking quality, and visualization confirms better user-specific semantic alignment.

Abstract

Retrieval-Augmented Generation (RAG) critically depends on effective query expansion to retrieve relevant information. However, existing expansion methods adopt uniform strategies that overlook user-specific semantics, ignoring individual expression styles, preferences, and historical context. In practice, identical queries in text can express vastly different intentions across users. This representational rigidity limits the ability of current RAG systems to generalize effectively in personalized settings. Specifically, we identify two core challenges for personalization: 1) user expression styles are inherently diverse, making it difficult for standard expansions to preserve personalized intent. 2) user corpora induce heterogeneous semantic structures-varying in topical focus and lexical organization-which hinders the effective anchoring of expanded queries within the user's corpora space. To address these challenges, we propose Personalize Before Retrieve (PBR), a framework that incorporates user-specific signals into query expansion prior to retrieval. PBR consists of two components: P-PRF, which generates stylistically aligned pseudo feedback using user history for simulating user expression style, and P-Anchor, which performs graph-based structure alignment over user corpora to capture its structure. Together, they produce personalized query representations tailored for retrieval. Experiments on two personalized benchmarks show that PBR consistently outperforms strong baselines, with up to 10% gains on PersonaBench across retrievers. Our findings demonstrate the value of modeling personalization before retrieval to close the semantic gap in user-adaptive RAG systems.

Personalize Before Retrieve: LLM-based Personalized Query Expansion for User-Centric Retrieval

TL;DR

The paper tackles the challenge of personalized query expansion in retrieval-augmented generation by proposing PBR, a framework that pre-emptively adapts queries using user-specific signals. It combines P-PRF, which generates style-aligned pseudo utterances and reasoning, with P-Anchor, which grounds queries in a semantic graph of the user’s private corpus via Personalized PageRank. The fusion step dynamically weights these signals to produce a personalized query representation, leading to improved retrieval performance on PersonaBench and LongMemEval across multiple retrievers. Experiments and ablations demonstrate that pre-retrieval personalization yields substantial gains in both recall and ranking quality, and visualization confirms better user-specific semantic alignment.

Abstract

Retrieval-Augmented Generation (RAG) critically depends on effective query expansion to retrieve relevant information. However, existing expansion methods adopt uniform strategies that overlook user-specific semantics, ignoring individual expression styles, preferences, and historical context. In practice, identical queries in text can express vastly different intentions across users. This representational rigidity limits the ability of current RAG systems to generalize effectively in personalized settings. Specifically, we identify two core challenges for personalization: 1) user expression styles are inherently diverse, making it difficult for standard expansions to preserve personalized intent. 2) user corpora induce heterogeneous semantic structures-varying in topical focus and lexical organization-which hinders the effective anchoring of expanded queries within the user's corpora space. To address these challenges, we propose Personalize Before Retrieve (PBR), a framework that incorporates user-specific signals into query expansion prior to retrieval. PBR consists of two components: P-PRF, which generates stylistically aligned pseudo feedback using user history for simulating user expression style, and P-Anchor, which performs graph-based structure alignment over user corpora to capture its structure. Together, they produce personalized query representations tailored for retrieval. Experiments on two personalized benchmarks show that PBR consistently outperforms strong baselines, with up to 10% gains on PersonaBench across retrievers. Our findings demonstrate the value of modeling personalization before retrieval to close the semantic gap in user-adaptive RAG systems.

Paper Structure

This paper contains 32 sections, 14 equations, 6 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Problem Definition
  3. PBR Framework
  4. Framework Overview
  5. P-PRF
  6. Pseudo Utterance Generation (Roughly)
  7. Pseudo Reasoning Generation (Logically)
  8. P-Anchor
  9. Graph Construction from User Corpus.
  10. Graph-Based Anchor Representation.
  11. PBR Fusion
  12. Final Retrieval.
  13. Experiment
  14. Experiment Setting
  15. Datasets.
  16. ...and 17 more sections

Figures (6)

  • Figure 1: An example comparing generic and personalized query expansion.
  • Figure 2: Overview of PBR, which consists of three components: P-PRF for stylistic expansion, P-Anchor for structural grounding, and PBR Fusion module for generating the final query representation.
  • Figure 3: Overall performance in PersonaBench comparison across three retrieval models.
  • Figure 4: Parameter sensitivity analysis on P-Anchor of $\theta$ and $k_2$ settings.
  • Figure 5: Visualization of retrieval distance of the query in the Figure \ref{['fig:intro']} compared to ground truth (GT) under different methods: (a) vanilla query, (b) HyDE, and (c) our proposed PBR.
  • ...and 1 more figures