Learning to Shop Like Humans: A Review-driven Retrieval-Augmented Recommendation Framework with LLMs Article Swipe

PDF

Kaiwen Wei , J. Gao , Jiang Zhong , Yuming Yang , Fangfang Lv , Zhihong Li ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2509.00698

Large language models (LLMs) have shown strong potential in recommendation tasks due to their strengths in language understanding, reasoning and knowledge integration. These capabilities are especially beneficial for review-based recommendation, which relies on semantically rich user-generated texts to reveal fine-grained user preferences and item attributes. However, effectively incorporating reviews into LLM-based recommendation remains challenging due to (1) inefficient to dynamically utilize user reviews under LLMs' constrained context windows, and (2) lacking effective mechanisms to prioritize reviews most relevant to the user's current decision context. To address these challenges, we propose RevBrowse, a review-driven recommendation framework inspired by the "browse-then-decide" decision process commonly observed in online user behavior. RevBrowse integrates user reviews into the LLM-based reranking process to enhance its ability to distinguish between candidate items. To improve the relevance and efficiency of review usage, we introduce PrefRAG, a retrieval-augmented module that disentangles user and item representations into structured forms and adaptively retrieves preference-relevant content conditioned on the target item. Extensive experiments on four Amazon review datasets demonstrate that RevBrowse achieves consistent and significant improvements over strong baselines, highlighting its generalizability and effectiveness in modeling dynamic user preferences. Furthermore, since the retrieval-augmented process is transparent, RevBrowse offers a certain level of interpretability by making visible which reviews influence the final recommendation.

Related Topics

Learning Theory (Education)

Life Or Something Like It

Dead Like Me

Attention (Machine Learning)

Concepts

No concepts available.

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2509.00698
PDF: https://arxiv.org/pdf/2509.00698
OA Status: green
OpenAlex ID: https://openalex.org/W4416685084

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4416685084

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2509.00698

Digital Object Identifier
Title: Learning to Shop Like Humans: A Review-driven Retrieval-Augmented Recommendation Framework with LLMs

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2025

Year of publication
Publication date: 2025-08-31

Full publication date if available
Authors: Kaiwen Wei, J. Gao, Jiang Zhong, Yuming Yang, Fangfang Lv, Zhihong Li

List of authors in order
Landing page: https://arxiv.org/abs/2509.00698

Publisher landing page
PDF URL: https://arxiv.org/pdf/2509.00698

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2509.00698

Direct OA link when available
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W4416685084
doi	https://doi.org/10.48550/arxiv.2509.00698
ids.doi	https://doi.org/10.48550/arxiv.2509.00698
ids.openalex	https://openalex.org/W4416685084
fwci
type	preprint
title	Learning to Shop Like Humans: A Review-driven Retrieval-Augmented Recommendation Framework with LLMs
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
is_xpac	False
apc_list
apc_paid
language	en
locations[0].id	pmh:oai:arXiv.org:2509.00698
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2509.00698
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2509.00698
locations[1].id	doi:10.48550/arxiv.2509.00698
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2509.00698
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5068039769
authorships[0].author.orcid	https://orcid.org/0000-0002-5830-0802
authorships[0].author.display_name	Kaiwen Wei
authorships[0].author_position	first
authorships[0].raw_author_name	Wei, Kaiwen
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5073488986
authorships[1].author.orcid	https://orcid.org/0000-0002-0193-9592
authorships[1].author.display_name	J. Gao
authorships[1].author_position	middle
authorships[1].raw_author_name	Gao, Jinpeng
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5120429402
authorships[2].author.orcid
authorships[2].author.display_name	Jiang Zhong
authorships[2].author_position	middle
authorships[2].raw_author_name	Zhong, Jiang
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5100958962
authorships[3].author.orcid
authorships[3].author.display_name	Yuming Yang
authorships[3].author_position	middle
authorships[3].raw_author_name	Yang, Yuming
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5040983343
authorships[4].author.orcid	https://orcid.org/0000-0001-6636-3056
authorships[4].author.display_name	Fangfang Lv
authorships[4].author_position	middle
authorships[4].raw_author_name	Lv, Fengmao
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5120391994
authorships[5].author.orcid	https://orcid.org/0000-0002-5428-2053
authorships[5].author.display_name	Zhihong Li
authorships[5].author_position	last
authorships[5].raw_author_name	Li, Zhenyang
authorships[5].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2509.00698
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	Learning to Shop Like Humans: A Review-driven Retrieval-Augmented Recommendation Framework with LLMs
has_fulltext	False
is_retracted	False
updated_date	2025-11-28T20:38:47.702120
primary_topic
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2509.00698
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2509.00698
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2509.00698
primary_location.id	pmh:oai:arXiv.org:2509.00698
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2509.00698
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2509.00698
publication_date	2025-08-31
publication_year	2025
referenced_works_count	0
abstract_inverted_index.a	91, 137, 196
abstract_inverted_index.To	84, 125
abstract_inverted_index.by	96, 201
abstract_inverted_index.in	8, 15, 103, 182
abstract_inverted_index.is	192
abstract_inverted_index.of	131, 199
abstract_inverted_index.on	32, 155, 161
abstract_inverted_index.to	12, 37, 55, 58, 73, 78, 116, 120
abstract_inverted_index.we	88, 134
abstract_inverted_index.(1)	56
abstract_inverted_index.(2)	69
abstract_inverted_index.and	19, 42, 68, 129, 143, 149, 171, 180
abstract_inverted_index.are	24
abstract_inverted_index.due	11, 54
abstract_inverted_index.for	27
abstract_inverted_index.its	118, 178
abstract_inverted_index.the	79, 97, 112, 127, 156, 189, 207
abstract_inverted_index.four	162
abstract_inverted_index.have	4
abstract_inverted_index.into	49, 111, 146
abstract_inverted_index.item	43, 144
abstract_inverted_index.most	76
abstract_inverted_index.over	174
abstract_inverted_index.rich	34
abstract_inverted_index.that	140, 167
abstract_inverted_index.user	40, 61, 105, 109, 142, 185
abstract_inverted_index.LLMs'	64
abstract_inverted_index.Large	0
abstract_inverted_index.These	22
abstract_inverted_index.final	208
abstract_inverted_index.forms	148
abstract_inverted_index.item.	158
abstract_inverted_index.level	198
abstract_inverted_index.shown	5
abstract_inverted_index.since	188
abstract_inverted_index.tasks	10
abstract_inverted_index.texts	36
abstract_inverted_index.their	13
abstract_inverted_index.these	86
abstract_inverted_index.under	63
abstract_inverted_index.which	30, 204
abstract_inverted_index.(LLMs)	3
abstract_inverted_index.Amazon	163
abstract_inverted_index.items.	124
abstract_inverted_index.making	202
abstract_inverted_index.models	2
abstract_inverted_index.module	139
abstract_inverted_index.offers	195
abstract_inverted_index.online	104
abstract_inverted_index.relies	31
abstract_inverted_index.reveal	38
abstract_inverted_index.review	132, 164
abstract_inverted_index.strong	6, 175
abstract_inverted_index.target	157
abstract_inverted_index.usage,	133
abstract_inverted_index.user's	80
abstract_inverted_index.ability	119
abstract_inverted_index.address	85
abstract_inverted_index.between	122
abstract_inverted_index.certain	197
abstract_inverted_index.content	153
abstract_inverted_index.context	66
abstract_inverted_index.current	81
abstract_inverted_index.dynamic	184
abstract_inverted_index.enhance	117
abstract_inverted_index.improve	126
abstract_inverted_index.lacking	70
abstract_inverted_index.process	100, 115, 191
abstract_inverted_index.propose	89
abstract_inverted_index.remains	52
abstract_inverted_index.reviews	48, 62, 75, 110, 205
abstract_inverted_index.utilize	60
abstract_inverted_index.visible	203
abstract_inverted_index.However,	45
abstract_inverted_index.PrefRAG,	136
abstract_inverted_index.achieves	169
abstract_inverted_index.commonly	101
abstract_inverted_index.context.	83
abstract_inverted_index.datasets	165
abstract_inverted_index.decision	82, 99
abstract_inverted_index.inspired	95
abstract_inverted_index.language	1, 16
abstract_inverted_index.modeling	183
abstract_inverted_index.observed	102
abstract_inverted_index.relevant	77
abstract_inverted_index.windows,	67
abstract_inverted_index.Extensive	159
abstract_inverted_index.LLM-based	50, 113
abstract_inverted_index.RevBrowse	107, 168, 194
abstract_inverted_index.behavior.	106
abstract_inverted_index.candidate	123
abstract_inverted_index.effective	71
abstract_inverted_index.framework	94
abstract_inverted_index.influence	206
abstract_inverted_index.introduce	135
abstract_inverted_index.knowledge	20
abstract_inverted_index.potential	7
abstract_inverted_index.reasoning	18
abstract_inverted_index.relevance	128
abstract_inverted_index.reranking	114
abstract_inverted_index.retrieves	151
abstract_inverted_index.strengths	14
abstract_inverted_index.RevBrowse,	90
abstract_inverted_index.adaptively	150
abstract_inverted_index.baselines,	176
abstract_inverted_index.beneficial	26
abstract_inverted_index.consistent	170
abstract_inverted_index.efficiency	130
abstract_inverted_index.especially	25
abstract_inverted_index.integrates	108
abstract_inverted_index.mechanisms	72
abstract_inverted_index.prioritize	74
abstract_inverted_index.structured	147
abstract_inverted_index.attributes.	44
abstract_inverted_index.challenges,	87
abstract_inverted_index.challenging	53
abstract_inverted_index.conditioned	154
abstract_inverted_index.constrained	65
abstract_inverted_index.demonstrate	166
abstract_inverted_index.distinguish	121
abstract_inverted_index.dynamically	59
abstract_inverted_index.effectively	46
abstract_inverted_index.experiments	160
abstract_inverted_index.inefficient	57
abstract_inverted_index.preferences	41
abstract_inverted_index.significant	172
abstract_inverted_index.Furthermore,	187
abstract_inverted_index.capabilities	23
abstract_inverted_index.disentangles	141
abstract_inverted_index.fine-grained	39
abstract_inverted_index.highlighting	177
abstract_inverted_index.improvements	173
abstract_inverted_index.integration.	21
abstract_inverted_index.preferences.	186
abstract_inverted_index.review-based	28
abstract_inverted_index.semantically	33
abstract_inverted_index.transparent,	193
abstract_inverted_index.effectiveness	181
abstract_inverted_index.incorporating	47
abstract_inverted_index.review-driven	92
abstract_inverted_index.recommendation	9, 51, 93
abstract_inverted_index.understanding,	17
abstract_inverted_index.user-generated	35
abstract_inverted_index.recommendation,	29
abstract_inverted_index.recommendation.	209
abstract_inverted_index.representations	145
abstract_inverted_index.generalizability	179
abstract_inverted_index.interpretability	200
abstract_inverted_index.preference-relevant	152
abstract_inverted_index.retrieval-augmented	138, 190
abstract_inverted_index."browse-then-decide"	98
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	6
citation_normalized_percentile