Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

One for Dozens: Adaptive REcommendation for All Domains with Counterfactual Augmentation

Huishi Luo, Yiwen Chen, Yiqing Wu, Fuzhen Zhuang, Deqing Wang

TL;DR

This paper addresses multi-domain recommendation with dozens of domains by introducing AREAD, a framework that combines a Hierarchical Expert Integration (HEI) with Hierarchical Expert Mask Pruning (HEMP) and a popularity-based counterfactual augmenter. HEI enables cross-domain knowledge sharing across multiple granularities using a compact hierarchical network, while HEMP learns domain-specific masks to identify effective transfers, guided by the Lottery Ticket Hypothesis. The counterfactual augmenter augments data for minor domains by leveraging cross-domain genuine-interest signals from popular items, mitigating data sparsity. Across two public datasets with over twenty domains, AREAD delivers consistent gains, especially for minor domains, while maintaining reasonable training and maintenance costs and providing interpretable insights into domain relationships and expert usage.

Abstract

Multi-domain recommendation (MDR) aims to enhance recommendation performance across various domains. However, real-world recommender systems in online platforms often need to handle dozens or even hundreds of domains, far exceeding the capabilities of traditional MDR algorithms, which typically focus on fewer than five domains. Key challenges include a substantial increase in parameter count, high maintenance costs, and intricate knowledge transfer patterns across domains. Furthermore, minor domains often suffer from data sparsity, leading to inadequate training in classical methods. To address these issues, we propose Adaptive REcommendation for All Domains with counterfactual augmentation (AREAD). AREAD employs a hierarchical structure with a limited number of expert networks at several layers, to effectively capture domain knowledge at different granularities. To adaptively capture the knowledge transfer pattern across domains, we generate and iteratively prune a hierarchical expert network selection mask for each domain during training. Additionally, counterfactual assumptions are used to augment data in minor domains, supporting their iterative mask pruning. Our experiments on two public datasets, each encompassing over twenty domains, demonstrate AREAD's effectiveness, especially in data-sparse domains. Source code is available at https://github.com/Chrissie-Law/AREAD-Multi-Domain-Recommendation.

One for Dozens: Adaptive REcommendation for All Domains with Counterfactual Augmentation

TL;DR

This paper addresses multi-domain recommendation with dozens of domains by introducing AREAD, a framework that combines a Hierarchical Expert Integration (HEI) with Hierarchical Expert Mask Pruning (HEMP) and a popularity-based counterfactual augmenter. HEI enables cross-domain knowledge sharing across multiple granularities using a compact hierarchical network, while HEMP learns domain-specific masks to identify effective transfers, guided by the Lottery Ticket Hypothesis. The counterfactual augmenter augments data for minor domains by leveraging cross-domain genuine-interest signals from popular items, mitigating data sparsity. Across two public datasets with over twenty domains, AREAD delivers consistent gains, especially for minor domains, while maintaining reasonable training and maintenance costs and providing interpretable insights into domain relationships and expert usage.

Abstract

Multi-domain recommendation (MDR) aims to enhance recommendation performance across various domains. However, real-world recommender systems in online platforms often need to handle dozens or even hundreds of domains, far exceeding the capabilities of traditional MDR algorithms, which typically focus on fewer than five domains. Key challenges include a substantial increase in parameter count, high maintenance costs, and intricate knowledge transfer patterns across domains. Furthermore, minor domains often suffer from data sparsity, leading to inadequate training in classical methods. To address these issues, we propose Adaptive REcommendation for All Domains with counterfactual augmentation (AREAD). AREAD employs a hierarchical structure with a limited number of expert networks at several layers, to effectively capture domain knowledge at different granularities. To adaptively capture the knowledge transfer pattern across domains, we generate and iteratively prune a hierarchical expert network selection mask for each domain during training. Additionally, counterfactual assumptions are used to augment data in minor domains, supporting their iterative mask pruning. Our experiments on two public datasets, each encompassing over twenty domains, demonstrate AREAD's effectiveness, especially in data-sparse domains. Source code is available at https://github.com/Chrissie-Law/AREAD-Multi-Domain-Recommendation.

Paper Structure

This paper contains 25 sections, 1 theorem, 9 equations, 8 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries and Background
  4. Overall Framework
  5. Hierarchical Expert Integration
  6. Why HEI
  7. Main Description
  8. Hierarchical Expert Mask Pruning
  9. Why HEMP
  10. Main Description
  11. Popularity-based Counterfactual Augmentation
  12. Model Optimization
  13. Experiments
  14. Experimental Setup
  15. Datasets
  16. ...and 10 more sections

Key Result

Corollary 1

If a user $u$ has a positive interaction $y=1$ with a unpopular item $i$ in a major domain, then it is also likely to occur in a minor domain: where $\mathcal{D}_a$ represents the set of major domains and $\mathcal{D}_b$ represents the set of minor domains. $p(i)$ means the popularity of $i$ correspond to $\bm{x}$ and $\rho$ is a certain popularity threshold.

Figures (8)

  • Figure 1: Sample size across 25 Amazon dataset domains, with 12 minor comprising less than 2% of total samples.
  • Figure 2: Hierarchical Expert Integration (HEI) uses multi-layer expert networks atop a base recommender to extract and integrate domain knowledge of varying granularities.
  • Figure 3: Hierarchical Expert Mask Pruning (HEMP) generates and iteratively prunes to select domain-specific experts.
  • Figure 4: Popularity-based Counterfactual Augmenter utilizes counterfactual reasoning to infer genuine user interest across domains, augmenting interactions with unpopular items in major domains to minor domains.
  • Figure 5: Comparison of performance, storage space, and training time across Single-domain, Multi-domain models, Isolated method, and AREAD on the Amazon dataset.
  • ...and 3 more figures

Theorems & Definitions (1)

  • Corollary 1