Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Centralized-Distributed Transfer Model for Cross-Domain Recommendation Based on Multi-Source Heterogeneous Transfer Learning

Ke Xu, Ziliang Wang, Wei Zheng, Yuhao Ma, Chenglin Wang, Nengxue Jiang, Cai Cao

TL;DR

A centralized-distributed transfer model (CDTM) for CDR based on multi-source heterogeneous transfer learning is proposed, which builds a dual embedding structure: domain specific embedding and global shared embedding to model the feature representation in the single domain and the commonalities in the global space, separately.

Abstract

Cross-domain recommendation (CDR) methods are proposed to tackle the sparsity problem in click through rate (CTR) estimation. Existing CDR methods directly transfer knowledge from the source domains to the target domain and ignore the heterogeneities among domains, including feature dimensional heterogeneity and latent space heterogeneity, which may lead to negative transfer. Besides, most of the existing methods are based on single-source transfer, which cannot simultaneously utilize knowledge from multiple source domains to further improve the model performance in the target domain. In this paper, we propose a centralized-distributed transfer model (CDTM) for CDR based on multi-source heterogeneous transfer learning. To address the issue of feature dimension heterogeneity, we build a dual embedding structure: domain specific embedding (DSE) and global shared embedding (GSE) to model the feature representation in the single domain and the commonalities in the global space,separately. To solve the latent space heterogeneity, the transfer matrix and attention mechanism are used to map and combine DSE and GSE adaptively. Extensive offline and online experiments demonstrate the effectiveness of our model.

A Centralized-Distributed Transfer Model for Cross-Domain Recommendation Based on Multi-Source Heterogeneous Transfer Learning

TL;DR

A centralized-distributed transfer model (CDTM) for CDR based on multi-source heterogeneous transfer learning is proposed, which builds a dual embedding structure: domain specific embedding and global shared embedding to model the feature representation in the single domain and the commonalities in the global space, separately.

Abstract

Cross-domain recommendation (CDR) methods are proposed to tackle the sparsity problem in click through rate (CTR) estimation. Existing CDR methods directly transfer knowledge from the source domains to the target domain and ignore the heterogeneities among domains, including feature dimensional heterogeneity and latent space heterogeneity, which may lead to negative transfer. Besides, most of the existing methods are based on single-source transfer, which cannot simultaneously utilize knowledge from multiple source domains to further improve the model performance in the target domain. In this paper, we propose a centralized-distributed transfer model (CDTM) for CDR based on multi-source heterogeneous transfer learning. To address the issue of feature dimension heterogeneity, we build a dual embedding structure: domain specific embedding (DSE) and global shared embedding (GSE) to model the feature representation in the single domain and the commonalities in the global space,separately. To solve the latent space heterogeneity, the transfer matrix and attention mechanism are used to map and combine DSE and GSE adaptively. Extensive offline and online experiments demonstrate the effectiveness of our model.

Paper Structure

This paper contains 27 sections, 9 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. CTR Estimation
  4. Cross Domain Recommendation
  5. Model Description
  6. Architecture Overview
  7. Embedding Layer
  8. Dual Embedding Structure
  9. Combination Layer
  10. Transfer Matrix
  11. Combination Attention
  12. Deep Layer and Output Layer
  13. Model Training and Auxiliary Loss
  14. Symmetrical Structure and Extensibility
  15. Experiments
  16. ...and 12 more sections

Figures (2)

  • Figure 1: The Framework of proposed Centralized-Distributed Transfer Model (GSE ie. Global shared embedding, DSE ie. Domain specific embedding. The Deep neural networks for each domain can be any single-domain model, such as DeepFM,DIN,etc.)
  • Figure 2: Comparison of ablation experiments (DA represents the model that removes combination attention, and ALL is the original model)