arxiv_ai 95% Match Research Paper GNN Researchers,Sequence Modeling Researchers,Data Scientists,Machine Learning Engineers 1 week ago

Bridging the Divide: End-to-End Sequence-Graph Learning

graph-neural-networks › graph-learning

📄 Abstract

Abstract: Many real-world datasets are both sequential and relational: each node carries an event sequence while edges encode interactions. Existing methods in sequence modeling and graph modeling often neglect one modality or the other. We argue that sequences and graphs are not separate problems but complementary facets of the same dataset, and should be learned jointly. We introduce BRIDGE, a unified end-to-end architecture that couples a sequence encoder with a GNN under a single objective, allowing gradients to flow across both modules and learning task-aligned representations. To enable fine-grained token-level message passing among neighbors, we add TOKENXATTN, a token-level cross-attention layer that passes messages between events in neighboring sequences. Across two settings, friendship prediction (Brightkite) and fraud detection (Amazon), BRIDGE consistently outperforms static GNNs, temporal graph methods, and sequence-only baselines on ranking and classification metrics.

Authors (8)

Yuen Chen

Yulun Wu

Samuel Sharpe

Igor Melnyk

Nam H. Nguyen

Furong Huang

+2 more

Submitted

October 29, 2025

arXiv Category

cs.LG

arXiv PDF

Key Contributions

Introduces BRIDGE, a unified end-to-end architecture that jointly learns from sequential and relational (graph) data. It couples a sequence encoder with a GNN under a single objective and incorporates TOKENXATTN for fine-grained message passing between events in neighboring sequences, enabling gradients to flow across both modalities.

Business Value

Enables more accurate predictions and insights from complex datasets that exhibit both temporal and relational structures, improving applications like targeted advertising, risk assessment, and user behavior analysis.

Paper Metadata

Innovation Type

Architectural Innovation

Deployment Feasibility

Moderate. Requires careful implementation of the combined architecture. Performance depends on the quality and nature of the sequence-graph data.

Limitations Addressed

Existing methods neglecting either sequential or graph modalities,Difficulty in jointly learning from both sequence and graph data,Lack of fine-grained interaction between events in neighboring sequences

Performance Gains

Consistently outperforms static GNNs, temporal graph methods, and sequence-only baselines.

Technical Tags

sequence modelinggraph modelingend-to-end learningunified architecturetoken-level cross-attentionGNNtemporal graphsrelational data

Research Topics

Graph Neural NetworksSequence ModelingRepresentation LearningMachine LearningData Mining

Methods & Architectures

BRIDGE architectureSequence EncoderGNNTOKENXATTN (token-level cross-attention) BRIDGE (unified sequence-graph architecture)

Applications & Tasks

Social Network Analysis Fraud Detection Recommendation Systems Cybersecurity Time Series Analysis Learning from sequential and relational dataJoint modeling of sequences and graphsNode classificationLink prediction Friendship predictionFraud detectionLearning task-aligned representations from sequence-graph data

Datasets & Benchmarks

Datasets

Brightkite, Amazon

Ranking metricsClassification metrics

Related Fields

Machine LearningGraph Neural NetworksSequence ModelingData MiningArtificial Intelligence

Keywords

Sequence ModelingGraph Neural NetworksBRIDGETOKENXATTNEnd-to-End LearningRepresentation LearningRelational DataTemporal GraphsFraud DetectionFriendship PredictionCross-AttentionUnified Architecture

Academic Context

#Graph Neural Networks#Sequence Modeling#Representation Learning#Machine Learning#Data Mining

Commercial Potential

Potential Products

Advanced fraud detection systemsSmarter recommendation enginesSocial network analysis tools

Target Industries

FinanceE-commerceSocial MediaCybersecurityTelecommunications

Use Case Examples

Detecting fraudulent transactions by analyzing user activity sequences and their social connections.Predicting future friendships in social networks based on interaction histories and network structure.Modeling user behavior in online platforms.

Competitive Edge

Provides a unified approach to modeling data that is inherently both sequential and relational, overcoming limitations of methods that treat these aspects separately.

Market Opportunity

Large market for AI solutions that can process complex, multi-modal data.

Revenue Models

Licensing of the BRIDGE frameworkintegration into data analytics platformsconsulting services.

Resource Requirements

Compute Needs

High (training combined sequence and graph models)

Data Requirements

Datasets containing both event sequences and graph structures (e.g., user activity logs with social connections).

Deployment Constraints

Computational cost, data preprocessing complexity, need for both sequence and graph data.

Scalability

Scalability depends on the underlying GNN and sequence encoder. Cross-attention might add computational overhead.

Regulatory Considerations

Data privacyespecially when dealing with user activity and social connections.

Production Readiness

Maturity Level

Research/Development

Time to Market

1-3 years

Patent Potential

Moderate, for the BRIDGE architecture and TOKENXATTN mechanism.

View Full Paper Back to Papers