Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Domain Latent Factors Sharing via Implicit Matrix Factorization

Abdulaziz Samra, Evgeney Frolov, Alexey Vasilev, Alexander Grigorievskiy, Anton Vakhrushev

TL;DR

The CDIMF is introduced, a model that extends the standard implicit matrix factorization with ALS to cross-domain scenarios and applies the Alternating Direction Method of Multipliers to learn shared latent factors for overlapped users while factorizing the interaction matrix.

Abstract

Data sparsity has been one of the long-standing problems for recommender systems. One of the solutions to mitigate this issue is to exploit knowledge available in other source domains. However, many cross-domain recommender systems introduce a complex architecture that makes them less scalable in practice. On the other hand, matrix factorization methods are still considered to be strong baselines for single-domain recommendations. In this paper, we introduce the CDIMF, a model that extends the standard implicit matrix factorization with ALS to cross-domain scenarios. We apply the Alternating Direction Method of Multipliers to learn shared latent factors for overlapped users while factorizing the interaction matrix. In a dual-domain setting, experiments on industrial datasets demonstrate a competing performance of CDIMF for both cold-start and warm-start. The proposed model can outperform most other recent cross-domain and single-domain models. We also provide the code to reproduce experiments on GitHub.

Cross-Domain Latent Factors Sharing via Implicit Matrix Factorization

TL;DR

The CDIMF is introduced, a model that extends the standard implicit matrix factorization with ALS to cross-domain scenarios and applies the Alternating Direction Method of Multipliers to learn shared latent factors for overlapped users while factorizing the interaction matrix.

Abstract

Data sparsity has been one of the long-standing problems for recommender systems. One of the solutions to mitigate this issue is to exploit knowledge available in other source domains. However, many cross-domain recommender systems introduce a complex architecture that makes them less scalable in practice. On the other hand, matrix factorization methods are still considered to be strong baselines for single-domain recommendations. In this paper, we introduce the CDIMF, a model that extends the standard implicit matrix factorization with ALS to cross-domain scenarios. We apply the Alternating Direction Method of Multipliers to learn shared latent factors for overlapped users while factorizing the interaction matrix. In a dual-domain setting, experiments on industrial datasets demonstrate a competing performance of CDIMF for both cold-start and warm-start. The proposed model can outperform most other recent cross-domain and single-domain models. We also provide the code to reproduce experiments on GitHub.
Paper Structure (26 sections, 26 equations, 4 figures, 3 tables, 1 algorithm)

This paper contains 26 sections, 26 equations, 4 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Related Work
  4. Proposed Model
  5. Update Iterations for CDIMF
  6. Proximal Operator Options
  7. Identity Operator
  8. L2 Regularization
  9. Details of the used ALS solver
  10. Derivation of equations for the iterations of the ALS solver
  11. Algorithm and Complexity
  12. Experimental Setup
  13. Datasets
  14. Evaluation Method and Metrics
  15. Baselines
  16. ...and 11 more sections

Figures (4)

  • Figure 1: Train/test data splits for (a) intra-domain recommendation scheme (warm start) and (b) inter-domain recommendation scheme (cold start); circles refer to training interactions, and squares refer to test interactions. Arrows refer to latent factors transfer for testing
  • Figure 2: HR@10, NDCG@10, and coverage changes in the warm start problem for several values of sharing parameter $\rho$. For $\rho=0$, a single domain ALS is running on each side.
  • Figure 3: The effect of L2 regularization on performance of CDIMF. Scenario: warm start, dataset: Cloth&Sport
  • Figure 4: HR@10 and NDCG@10 changes in the warm start problem for several values of aggregation period (AP), squares refer to epochs of shared variable update