Blending Sequential Embeddings, Graphs, and Engineered Features: 4th Place Solution in RecSys Challenge 2025

2025-08-09

AI Summary

For the RecSys Challenge 2025 multi-task recommendation scenario, this work addresses the challenge of learning user embeddings with strong cross-task transferability across six diverse downstream tasks. Methodologically, we propose a unified multimodal representation framework that jointly models temporal interest evolution (via Transformer-based sequential encoding), structural relational patterns (via GNNs over user–item interaction graphs), and high-order feature interactions (via DCN augmented with task-aware feature engineering). Our key contribution lies in the holistic integration of temporal dynamics, graph-structured relationships, and nonlinear feature cross-effects—thereby significantly improving embedding generalizability and robustness. Empirically, the approach achieves fourth place in the RecSys Challenge 2025, demonstrating consistent and superior transfer performance across heterogeneous recommendation tasks.

Technology Category

• Machine Learning: Multimodal Learning
• Data Mining & Knowledge Management: Recommender Systems
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This paper describes the 4th-place solution by team ambitious for the RecSys Challenge 2025, organized by Synerise and ACM RecSys, which focused on universal behavioral modeling. The challenge objective was to generate user embeddings effective across six diverse downstream tasks. Our solution integrates (1) a sequential encoder to capture the temporal evolution of user interests, (2) a graph neural network to enhance generalization, (3) a deep cross network to model high-order feature interactions, and (4) performance-critical feature engineering.

Problem

• Generate effective user embeddings for diverse tasks
• Integrate sequential and graph models for user behavior
• Combine feature engineering with deep learning techniques

Innovation

• Sequential encoder captures temporal user interests
• Graph neural network enhances generalization
• Deep cross network models high-order interactions