Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Aggregator Time-Warping Heterogeneous Graph Neural Network for Personalized Micro-Video Recommendation

Jinkun Han, Wei Li, Zhipeng Cai, Yingshu Li

TL;DR

This work tackles micro-video personalization by introducing MTHGNN, a time-warping, multi-aggregator heterogeneous graph neural network that jointly models social relations, multi-modal content, and sequential, timeliness-driven user preferences. Key innovations include a directed multi-modal heterogeneous graph, relational and attention-based message passing, session-based attention with LSTM, and a graph-free sampling strategy for efficient inference. Empirical results on TikTok and MovieLens demonstrate consistent gains over strong baselines in precision, ranking, and timeliness, with ablations confirming the value of each component. The approach offers a practical pathway to more timely and context-aware micro-video recommendations in real-world systems.

Abstract

Micro-video recommendation is attracting global attention and becoming a popular daily service for people of all ages. Recently, Graph Neural Networks-based micro-video recommendation has displayed performance improvement for many kinds of recommendation tasks. However, the existing works fail to fully consider the characteristics of micro-videos, such as the high timeliness of news nature micro-video recommendation and sequential interactions of frequently changed interests. In this paper, a novel Multi-aggregator Time-warping Heterogeneous Graph Neural Network (MTHGNN) is proposed for personalized news nature micro-video recommendation based on sequential sessions, where characteristics of micro-videos are comprehensively studied, users' preference is mined via multi-aggregator, the temporal and dynamic changes of users' preference are captured, and timeliness is considered. Through the comparison with the state-of-the-arts, the experimental results validate the superiority of our MTHGNN model.

Multi-Aggregator Time-Warping Heterogeneous Graph Neural Network for Personalized Micro-Video Recommendation

TL;DR

This work tackles micro-video personalization by introducing MTHGNN, a time-warping, multi-aggregator heterogeneous graph neural network that jointly models social relations, multi-modal content, and sequential, timeliness-driven user preferences. Key innovations include a directed multi-modal heterogeneous graph, relational and attention-based message passing, session-based attention with LSTM, and a graph-free sampling strategy for efficient inference. Empirical results on TikTok and MovieLens demonstrate consistent gains over strong baselines in precision, ranking, and timeliness, with ablations confirming the value of each component. The approach offers a practical pathway to more timely and context-aware micro-video recommendations in real-world systems.

Abstract

Micro-video recommendation is attracting global attention and becoming a popular daily service for people of all ages. Recently, Graph Neural Networks-based micro-video recommendation has displayed performance improvement for many kinds of recommendation tasks. However, the existing works fail to fully consider the characteristics of micro-videos, such as the high timeliness of news nature micro-video recommendation and sequential interactions of frequently changed interests. In this paper, a novel Multi-aggregator Time-warping Heterogeneous Graph Neural Network (MTHGNN) is proposed for personalized news nature micro-video recommendation based on sequential sessions, where characteristics of micro-videos are comprehensively studied, users' preference is mined via multi-aggregator, the temporal and dynamic changes of users' preference are captured, and timeliness is considered. Through the comparison with the state-of-the-arts, the experimental results validate the superiority of our MTHGNN model.
Paper Structure (18 sections, 13 equations, 3 figures, 6 tables, 2 algorithms)

This paper contains 18 sections, 13 equations, 3 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Statement
  4. Methodology
  5. User-video Graph Structure and Data Processing
  6. Heterogeneous Graph Construction and Heterogeneous Graph Sessions
  7. Multi-aggregator
  8. Session Attention
  9. Sampling Strategy
  10. Model Training & Prediction
  11. Performance Evaluation
  12. Datasets
  13. Baseline Models
  14. Experiment Settings
  15. Comparison between MTHGNN and Baselines
  16. ...and 3 more sections

Figures (3)

  • Figure 1: The Framework of our proposed MTHGNN ($\oplus$ represents concatenation operation).
  • Figure 2: Homogeneous graph vs. multi-modal heterogeneous graph
  • Figure 3: Top-$K$ recommendation performance of MTHGNN on TikTok(1/50) ((a) Precision@K, (b) Revised Precision@K, (c) NDCG@K, (d) C-Timeliness@K)