Multi-TAP: Multi-criteria Target Adaptive Persona Modeling for Cross-Domain Recommendation

TL;DR

Experiments on real-world datasets demonstrate that Multi-TAP consistently outperforms state-of-the-art CDR methods, highlighting the importance of modeling intra-domain heterogeneity for robust cross-domain recommendation.

Abstract

Cross-domain recommendation (CDR) aims to alleviate data sparsity by transferring knowledge across domains, yet existing methods primarily rely on coarse-grained behavioral signals and often overlook intra-domain heterogeneity in user preferences. We propose Multi-TAP, a multi-criteria target-adaptive persona framework that explicitly captures such heterogeneity through semantic persona modeling. To enable effective transfer, Multi-TAP selectively incorporates source-domain signals conditioned on the target domain, preserving relevance during knowledge transfer. Experiments on real-world datasets demonstrate that Multi-TAP consistently outperforms state-of-the-art CDR methods, highlighting the importance of modeling intra-domain heterogeneity for robust cross-domain recommendation. The codebase of Multi-TAP is currently available at https://github.com/archivehee/Multi-TAP.

Figures (6)

• Figure 1: Preference shift ratios across price-based user groups within the Electronics domain. It reports the proportion of users whose relative price preference shifts when moving from a source category (Computer or Home Audio) to different target categories, averaged across target categories, using price as an example of a preference criterion.
• Figure 2: Intra-domain preference heterogeneity across categories. Each heatmap visualizes the conditional preference preservation ratio, measuring how consistently users maintain their dominant preference label when moving from a base category ($\boldsymbol{y}$-axis) to a compared category ($\boldsymbol{x}$-axis). Each cell represents the fraction of users who are assigned to a specific preference group (High or Low) in the base category and remain in the same group in the compared category. Lower values indicate weaker preference preservation and stronger intra-domain heterogeneity.
• Figure 3: Overview of Multi-TAP, which consists of (P1) Multi-criteria Persona Modeling, and (P2) Target-adaptive Doppelganger Transfer. In (P1), Multi-TAP decomposes user preferences into explicit, criterion-aware multi-persona embeddings by leveraging a user persona database constructed from interaction histories and item metadata. In (P2), Multi-TAP selectively incorporates source-domain information by instantiating doppelganger persona aligned with the target-domain persona embedding.
• Figure 4: Effect of doppelganger-based indirect transfer compared to direct alignment strategies in terms of HR@5.
• Figure 5: Sensitivity of Multi-TAP to the hyperparameters $\boldsymbol{\lambda}$ and $\boldsymbol{\tau}$ in terms of HR@5.