IntentGR: Enhancing Intent for End-to-End Personalized Generative Retrieval in Search

Gefan Ye, Yanjing Wu, Yinfu Feng, Zhiqiang Wang, Liang Li, Jun Xiao, Rong Xiao, Xiaoyi ZengarXiv 2026 · DOI · arXiv

TL;DR

IntentGR is a novel end-to-end generative retrieval framework that prevents "intent drift" in personalized search by reinforcing query semantics at every stage of the model pipeline. By explicitly aligning query intent with item identifiers and enriching long-tail query data, it achieves up to an 11.96% improvement in retrieval accuracy over state-of-the-art baselines.

What Problem Does This Paper Address?

Generative Retrieval (GR) is a rising paradigm where models directly "generate" the identifier (SID) of a relevant item, rather than ranking a pre-defined list. While end-to-end GR—which jointly learns how to tokenize items and how to retrieve them—has shown promise in recommendation systems, it often falters in personalized search.

The core issue is intent drift. In search, the model must balance a user's long-term preferences (e.g., "I like electronics") with their immediate intent (e.g., "I need a sleeping bag"). Current models often treat the search query as a passive input feature. Consequently, the generated identifiers tend to drift toward high-frequency token patterns (popular items) rather than the specific intent of the query. This is especially problematic for "long-tail" queries that appear infrequently in training data.

Figure 1: Comparison of three retrieval paradigms. Discrimi- native retrieval (a) ranks candidates by query-item matching scores; two-stage generative retrieval (b) first constructs item tokens and then trains a retriever; end-to-end generative re- trieval (c) jointly optimizes the tokenizer and retriever.

Methods

The authors propose IntentGR, which injects query intent into four complementary stages of the end-to-end pipeline:

  1. Query-Aware Semantic Behavior Representation (QBR): Beyond the user's personal history, QBR adds a "query-level behavior sequence"—a list of items globally popular for that specific query. This acts as a form of query-level collaborative filtering.
  2. Query-Semantic Alignment (QSA): This introduces a shared contrastive objective. It forces the model to align the representation of the query with the reconstructed representation of the target item, ensuring the learned "language" of identifiers is grounded in search intent.
  3. Long-Tail Query Enrichment (LQE): To combat the dominance of frequent queries, LQE uses repeated sampling and query rewriting (generating semantic variants of rare queries) to give the model more exposure to diverse, low-frequency intents.
  4. Query-conditioned Attention Modulation (QAM): During the actual generation of the item ID, QAM explicitly injects global query information into the decoder's attention mechanism. This ensures the model doesn't "forget" the query as it generates long sequences of ID tokens.
Figure 2: An overview of IntentGR. IntentGR consists of four key mechanisms: Query-Aware Semantic Behavior Representation (QBR), Query- Semantic Alignment (QSA), Long-Tail Query Enrichment (LQE), and Query-conditioned Attention Modulation (QAM). These mechanisms enhance intent from complementary perspectives, enabling IntentGR to better capture query intent and personalized preference in search.

The model follows an alternating training strategy, where it switches between optimizing the item tokenizer and the generative retriever before a final fine-tuning phase.

Figure 3: The training strategy of IntentGR, consisting of alternating training, where 𝑇𝜙and 𝑅𝜃are updated in turn, and GR fine-tuning, where 𝑇𝜙is frozen and 𝑅𝜃is further optimized.

Datasets

The framework was evaluated on two distinct datasets:

  • AOL4PS: A public benchmark for personalized search derived from AOL web search logs.
  • AE: A massive, real-world industrial dataset from AliExpress, containing over 11 million users and 9 million items. This dataset is particularly challenging due to the "lexical gap" between colloquial user queries and keyword-heavy product titles.

Key Insights and Ideas

  • Query-Level Evidence is a Game Changer: The authors found that simply adding query-level behavioral evidence (QBR) provides a much stronger signal for target items than user history alone, especially for inactive users.
  • Generalization vs. Memorization: By using QSA and QAM, the model moves from simply "memorizing" which items go with which queries to "abstracting" a semantic relationship. This allows the model to perform better even when it hasn't seen a specific user-query pair before.
  • Balancing the Tail: Long-tail queries represent the vast majority of unique intents (87% in AOL4PS). LQE proves that you don't need more data to improve these; you just need to utilize the existing "tail" data more effectively through augmentation and balanced supervision.

Key Results

IntentGR consistently outperformed both discriminative (like BM25) and generative (like TIGER and ETEGRec) baselines:

  • Recall@10 Improvements: It achieved an 8.11% gain on AOL4PS and a 11.96% gain on the industrial AE dataset compared to the strongest baseline.
  • Long-Tail Success: On the AE dataset, IntentGR improved Recall@10 for the "Low" frequency query bucket by a staggering amount, nearly doubling the performance of the baseline in some metrics.
  • Robustness: Hyper-parameter analysis showed that the model is relatively stable across different weights for its alignment and attention modulation components.

Limitations and Future Work

While IntentGR significantly improves intent-awareness, the authors note a few areas for growth:

  • Evidence Sources: The current model relies on text and click history. Future work could incorporate multimodal signals (like product images) or session-level context (the sequence of queries in a single sitting) to further refine intent.
  • Scalability: While tested on 9 million items, the computational cost of end-to-end training on even larger, rapidly changing corpora remains a challenge for all generative retrieval methods.

Summary updated

Abstract

Personalized generative retrieval in search should jointly satisfy the user's immediate search intent and long-term preference. Recent end-to-end approaches, primarily explored in recommendation, unify item tokenization and sequence generation within a single model, allowing semantic identifiers (SIDs) to be optimized toward the downstream generation objective. However, these methods often treat the query merely as an input feature, without explicitly leveraging it to guide either identifier learning or sequence generation. This leaves intent guidance weak in both the tokenization and the generation stage, causing generated semantic identifiers to drift toward high-frequency token patterns rather than the current search intent. In this work, we find that incorporating additional query-related evidence into the framework of end-to-end generative retrieval is important. Based on this finding, we propose IntentGR, an intent-enhanced end-to-end framework that strengthens query semantics at four complementary stages of the retrieval pipeline. Query-Aware Semantic Behavior Representation (QBR) introduces query-level behavioral evidence as a new input signal; Query-Semantic Alignment (QSA), Long-Tail Query Enrichment (LQE), and Query-conditioned Attention Modulation (QAM) then serve as integration mechanisms at the optimization, training-distribution, and decoding stages to fully exploit this evidence. Experiments on the public AOL4PS benchmark and AE, a large-scale industrial e-commerce dataset, show that IntentGR improves Recall@10 by 8.11% and 11.96% over the strongest baseline, with strong gains on long-tail queries. These results suggest that enriched query-related input evidence, together with effective integration mechanisms, is beneficial for building intent-aware end-to-end generative search systems.
BibTeX
@misc{ye2026intentgr,
  author = {Gefan Ye and Yanjing Wu and Yinfu Feng and Zhiqiang Wang and Liang Li and Jun Xiao and Rong Xiao and Xiaoyi Zeng},
  title = {{IntentGR: Enhancing Intent for End-to-End Personalized Generative Retrieval in Search}},
  year = {2026},
  eprint = {2611.22788},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.22788},
}
Figure