Cross-Domain Sequential Recommendation with Representation-Level User Transfer

Abdulaziz Samra; Nikita Sukhorukov; Alexey Grishanov; Alexey Vasilev; Evgeny Frolov

cs.IROct 26, 2026

Cross-Domain Sequential Recommendation with Representation-Level User Transfer

Authors: Abdulaziz Samra, Nikita Sukhorukov, Alexey Grishanov, Alexey Vasilev, Evgeny Frolov

Organizations: Skolkovo Institute of Science and Technology, Moscow, Russian Federation · Higher School of Economics, Moscow, Russian Federation · AXXX, Moscow, Russian Federation · Sber AI Lab, Moscow, Russian Federation

Abstract

Cross-domain sequential recommendation aims to improve next-item prediction in a target domain by leveraging user behavior from auxiliary source domains. However, real-world deployment faces two critical challenges: business constraints that isolate interaction data across domains, and user heterogeneity, where many users lack cross-domain activity. This makes existing cross-domain methods difficult to apply. In this paper, we propose a simple and practical framework for cross-domain sequential recommendations that transfers information exclusively through learned user representations, without sharing raw interaction data or item embeddings across domains. We introduce a lightweight adapter that projects the source-domain user representation into the target-domain embedding space and integrates it with the target interaction sequence. Importantly, when source data is unavailable (i.e., the user observed only in the target domain), the model seamlessly defaults to a robust target-only sequential recommender, retaining benefits from joint training across all users. Extensive experiments were conducted under two setups; one targets benchmarking against previous baselines, while the other focuses on realistic and scalable conditions. Both sets of experiments demonstrate consistent performance gains for shared and unshared users.

Figures & tables

Figure 1. The FLADRec architecture. The Source Domain (Green, upper) and Target Domain (Red, lower) streams process user sequences in parallel. Fusing Layer maps the source sequence representation $\mathbf{e}_{u}^{S}$ to the target space as $\mathbf{e}_{u}^{S,p}$ . This projected vector is then fused element-wise with the target item embeddings { $\mathbf{e}_{1}^{u,T},\dots,\mathbf{e}_{m}^{u,T}$ }, acting as a global user context. The fused embeddings { $\mathbf{e}_{\dots}^{u,fT}$ } are fed into the target encoder for final prediction $\mathbf{e}_{u}^{T}$ . Icons denote trainable (Flame) vs. frozen (Snowflake) parameters. Note: If a user has no source history, the fusion step is skipped.

Figure 2. Vertical federated learning protocol for cross-domain sequential recommendation. Throughout, raw interaction data and item IDs remain private to each domain; only embeddings and their gradients are exchanged.

Dataset	Users	Items	Inter.	Val Seq.	Test Seq.	Density
Food	7144	11837	83663	2837	2419	0.098%
Kitchen	7144	16258	89885	4307	4725	0.077%
Beauty	4474	10379	50329	2086	1875	0.108%
Electronics	4474	14188	63800	2388	2599	0.100%
Movie	28350	35712	347650	11728	10935	0.034%
Book	28350	90958	403147	16622	17415	0.015%

Table 1. Statistics of the CDSR benchmark datasets adopted from the same experimental setup from prior work ( Bian et al., 2025 ) .

Domains	Users	Items	Inter.	Shared	Overlap
QK-Article	449,288	180,951	14,887,703	224,644	50.0%
$\hookrightarrow$ QK-Video	449,288	694,609	11,784,281	224,644	50.0%
Zvuk	95,952	822,274	38,723,635	47,976	50.0%
$\hookrightarrow$ MegaMarket	95,952	453,367	4,754,096	47,976	50.0%

Table 2. Statistics of the large-scale CDSR datasets used in the scalability-oriented evaluation. We consider a unidirectional transfer setting, where the first domain in each pair serves as the source and the second as the target.

		Single Domain		Cross-Domain (Disjoint-data)
Domain	Metric	SASRec-1	BERT4Rec-1	CD-SASRec	FedDCSR	CD-ASR	FLADRec
Food	HR@5	0.1930	0.1819	0.1797	0.1337	0.1976	0.2303
	HR@10	0.2611	0.2528	0.2454	0.2274	0.2727	0.3043
	NDCG@10	0.1561	0.1462	0.1421	0.1214	0.1616	0.1851
	MRR	0.1332	0.1230	0.1197	0.0994	0.1368	0.1486
Kitchen	HR@5	0.1241	0.1114	0.1119	0.0987	0.1345	0.2241

Table 3. Benchmark results against models running under limited access to source interaction data

		Single Domain		Cross-Domain (Pooled-data)				Ours
Domain	Metric	SASRec-2	BERT4Rec-2	MGCL	C 2 DSR	DREAM	ABXI	FLADRec
Food	HR@5	0.2313	0.2223	0.1932	0.1984	0.2158	0.2498	0.2303
	HR@10	0.2854	0.2956	0.2673	0.2574	0.2771	0.3175	0.3043
	NDCG@10	0.1797	0.1727	0.1523	0.1546	0.1698	0.1973	0.1851
	MRR	0.1535	0.1427	0.1260	0.1311	0.1441	0.1679	0.1486
Kitchen	HR@5	0.1510	0.1363	0.1467	0.1263	0.1377	0.1737	0.2241

Table 4. Benchmark results against models running under full access to source interaction data

Dataset	Metric	SASRec	FLADRec (Ours)	Imp.
QK-article QK-video	NDCG@10	0.0225	0.0283	+24.1%
	NDCG@50	0.0382	0.0469	+22.8%
	NDCG@100	0.0469	0.0569	+21.3%
	HR@10	0.0423	0.0532	+25.8%
	HR@50	0.1155	0.1395	+20.8%
	HR@100	0.1691	0.2017	+19.3%

Table 5. Results of FLADRec model on large-scale datasets. NDCG, HitRate, Coverage (COV) at cutoff

K=10,50,100

and relative improvements w.r.t. the SASRec baseline are reported.

Dataset	ABXI	FLADRec			Speedup
		Pretrain	Adapt	Total
Food	42.0	0.39	0.86	3.63	11.57 $\times$
Kitchen		1.67	0.71
Beauty	17.0	1.18	0.94	3.00	5.67 $\times$
Electronics		0.33	0.55
Movie	65.6	5.29	2.45	17.82	3.68 $\times$

Table 6. Training time (minutes) comparison between ABXI and FLADRec under identical hardware conditions.

			User subset
Model	Regime	Metric	ALL	Shared	Unshared
SASRec	–	value	0.02248	0.00976	0.01272
FLADRec (Tuned Source)	full	value	0.02832	0.01231	0.01601
	full	gain,%	+26.0	+26.3	+25.8
	rand	value	0.02790	0.01226	0.01564
	rand	gain,%	+24.5	+25.7	+23.0

Table 7. NDCG@10 comparison across user subsets (ALL / Shared / Unshared). Relative gains w.r.t. baseline are reported.

Cross-Domain Recommendation

See all 12 papers

Breaking the Information Silo: Semantic Personas for Cross-Domain Recommendation
Jun 1, 2026Jonathan Mayo, Moshe Unger, Konstantin BaumanCross-Domain Recommendation Cross-Domain Transfer Learning
Technology and Information Management Department, Coller School of Management, Tel Aviv University, Tel Aviv 6997801, Israel · Management Information Systems Department, Fox School of Business, Temple University, Philadelphia, Pennsylvania 19122
CoRCi: Cross-Reconstruction of Coherent Interests Modeling in Cross-Domain Sequential Recommendation
Aug 10, 2026Qingtian Bian, Tieying Li, Marcus de Carvalho +3Sequential Recommendation Invariant Representation Learning
Nanyang Technological University Singapore · Northeastern University China · Shanghai University of Finance and Economics China
Empowering Cross-Domain Sequential Recommendation with Hybrid Tokenization and Serial-Parallel Decoding
Jul 21, 2026Yuxuan Hu, Yuhao Wang, Tianbo Huang +4Sequential Recommendation Parallel Decoding
City University of Hong Kong · Hong Kong, China · ByteDance Inc. +1

Jun 1, 2026J/K move · Enter open · S save

Breaking the Information Silo: Semantic Personas for Cross-Domain Recommendation

Jonathan Mayo, Moshe Unger, Konstantin Bauman

Technology and Information Management Department, Coller School of Management, Tel Aviv University, Tel Aviv 6997801, Israel · Management Information Systems Department, Fox School of Business, Temple University, Philadelphia, Pennsylvania 19122

Cross-Domain Recommendation Cross-Domain Transfer Learning Recommender Systems User Preference Modeling

Open paper AskPDF arXiv

Digital platforms increasingly operate as isolated information silos, limiting their ability to construct comprehensive user representations across domains. Cross-domain recommender systems seek to overcome this limitation by transferring knowledge from a source domain to a target domain, yet most existing approaches depend on shared users, shared items, or structurally similar interaction graphs. These assumptions are often unrealistic across independent platforms. We propose SPHERE (Semantic Personas for Heterogeneous cross-domain Recommendation), a design artifact that enables recommendation knowledge transfer across strictly disjoint domains with no shared users or items. Rather than aligning domains through identity or graph structure, SPHERE uses large language models to induce a shared behavioral vocabulary, generate structured semantic personas for users, and retrieve behaviorally similar source-domain communities that form a Community Source Persona. This semantic signal is integrated with collaborative signals through a dual-tower architecture and dynamic fusion gate, allowing SPHERE to augment standard recommender backbones. Empirical evaluation across Amazon Books, Goodreads, and Steam demonstrates consistent improvements over NCF, SVD++, and LightGCN baselines under full-ranking evaluation. The results show that cross-domain transfer effectiveness is not determined solely by semantic proximity between domains; rather, it depends critically on the structural density and native predictive strength of the target domain. The study contributes to information systems research by reframing cross-domain personalization as behavior-based semantic alignment, offering a practical mechanism for overcoming information silos while preserving interpretability and modularity.

Abstract

Cross-Domain Recommendation

Explore similar work

Breaking the Information Silo: Semantic Personas for Cross-Domain Recommendation

CoRCi: Cross-Reconstruction of Coherent Interests Modeling in Cross-Domain Sequential Recommendation

Empowering Cross-Domain Sequential Recommendation with Hybrid Tokenization and Serial-Parallel Decoding