Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Social Graph for Inactive User Recommendation

Nian Liu, Shen Fan, Ting Bai, Peng Wang, Mingwei Sun, Yanhu Mo, Xiaoxiao Xu, Hong Liu, Chuan Shi

TL;DR

This work addresses inactive-user recommendation where raw social graphs often lack sufficient quality and quantity. It introduces LSIR, which learns an optimal social graph by performing U2U deletion and U2C addition, guided by encoding user-item interactions and a mimic-learning mechanism that aligns active-user representations with inactive-interest clusters. The approach combines multi-hop embedding propagation, selective graph refinement, and a contrastive loss (InfoNCE) to improve inactive-user connectivity while preserving active-user performance, achieving significant gains (up to 129.58% in NDCG on some datasets). The results demonstrate the practicality and scalability of refining social structures in recommender systems, particularly for cold-start/inactive users on large industrial data.

Abstract

Social relations have been widely incorporated into recommender systems to alleviate data sparsity problem. However, raw social relations don't always benefit recommendation due to their inferior quality and insufficient quantity, especially for inactive users, whose interacted items are limited. In this paper, we propose a novel social recommendation method called LSIR (\textbf{L}earning \textbf{S}ocial Graph for \textbf{I}nactive User \textbf{R}ecommendation) that learns an optimal social graph structure for social recommendation, especially for inactive users. LSIR recursively aggregates user and item embeddings to collaboratively encode item and user features. Then, graph structure learning (GSL) is employed to refine the raw user-user social graph, by removing noisy edges and adding new edges based on the enhanced embeddings. Meanwhile, mimic learning is implemented to guide active users in mimicking inactive users during model training, which improves the construction of new edges for inactive users. Extensive experiments on real-world datasets demonstrate that LSIR achieves significant improvements of up to 129.58\% on NDCG in inactive user recommendation. Our code is available at~\url{https://github.com/liun-online/LSIR}.

Learning Social Graph for Inactive User Recommendation

TL;DR

This work addresses inactive-user recommendation where raw social graphs often lack sufficient quality and quantity. It introduces LSIR, which learns an optimal social graph by performing U2U deletion and U2C addition, guided by encoding user-item interactions and a mimic-learning mechanism that aligns active-user representations with inactive-interest clusters. The approach combines multi-hop embedding propagation, selective graph refinement, and a contrastive loss (InfoNCE) to improve inactive-user connectivity while preserving active-user performance, achieving significant gains (up to 129.58% in NDCG on some datasets). The results demonstrate the practicality and scalability of refining social structures in recommender systems, particularly for cold-start/inactive users on large industrial data.

Abstract

Social relations have been widely incorporated into recommender systems to alleviate data sparsity problem. However, raw social relations don't always benefit recommendation due to their inferior quality and insufficient quantity, especially for inactive users, whose interacted items are limited. In this paper, we propose a novel social recommendation method called LSIR (\textbf{L}earning \textbf{S}ocial Graph for \textbf{I}nactive User \textbf{R}ecommendation) that learns an optimal social graph structure for social recommendation, especially for inactive users. LSIR recursively aggregates user and item embeddings to collaboratively encode item and user features. Then, graph structure learning (GSL) is employed to refine the raw user-user social graph, by removing noisy edges and adding new edges based on the enhanced embeddings. Meanwhile, mimic learning is implemented to guide active users in mimicking inactive users during model training, which improves the construction of new edges for inactive users. Extensive experiments on real-world datasets demonstrate that LSIR achieves significant improvements of up to 129.58\% on NDCG in inactive user recommendation. Our code is available at~\url{https://github.com/liun-online/LSIR}.
Paper Structure (23 sections, 16 equations, 5 figures, 4 tables)

This paper contains 23 sections, 16 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Industrial Observations on Social Relation
  3. Preliminary
  4. The Proposed Model
  5. Encoding User-Item Interactions
  6. Graph Structure Learning on Social Graph
  7. U2U deletion
  8. U2C addition
  9. Mimic Learning
  10. Complexity Analysis
  11. Experiments
  12. Experimental Setup
  13. Overall Recommendation Performance(RQ1)
  14. Effects of Graph Structure Learning(RQ2)
  15. Analysis of GSL variants
  16. ...and 8 more sections

Figures (5)

  • Figure 1: The industrial observations on social relations based on Taobao.
  • Figure 2: The overview of our proposed LSIR, including 3 components: 1)Encoding User-Item Interactions, 2)Graph Structure Learning on Social Graph(U2U deletion & U2C addition) and 3)Mimic Learning.
  • Figure 3: Test on the Mimic Learning.
  • Figure 4: Test on the ratio of inactive users.
  • Figure 5: Test on the number of cluster $\mathcal{C}$.

Theorems & Definitions (2)

  • definition thmcounterdefinition
  • definition thmcounterdefinition