How truewangsyutung/ufgraphfr works
UFGraphFR is a federated recommendation research system. Its principal workflow is to load interaction and user-profile data, encode profile text, train client recommendation models across rounds, aggregate client parameters into a server-side user-relation graph, and evaluate ranked recommendations. The graph preserves the documented transformer and optional noise capabilities. Source excerpts show the data, embedding, model, and graph-building pieces, but do not expose the full round orchestration; uncertain links are omitted rather than inferred.
Main entry points: train.py Training Orchestrator · engine.py Federated Engine · mymodel.py UFGraphFR Client Model · metrics.py Ranking Metrics
Components 14
Data and Features
- Benchmark Data
data - Data Preparation
train.py - Interaction Sampling
data.py - Profile Text Encoding
embedding.py
Federated Learning
- Training Orchestrator
train.py - Federated Engine
engine.py - User Graph Utilities
utils.py - Optional Noise
engine.py
Recommendation Models
- UFGraphFR Client Model
mymodel.py - Sequence Transformer
mymodel.py - MLP Baseline
mlp.py
Evaluation
- Ranking Metrics
metrics.py - Evaluation Results
train.py
Other components
- Experiment Runner
Connections 14
- Experiment Runner → Training Orchestrator: runs
README.md - Training Orchestrator → Benchmark Data: loads
train.py - Training Orchestrator → Data Preparation: prepares
train.py - Data Preparation → Interaction Sampling: uses
train.py - Training Orchestrator → Profile Text Encoding: configures
train.py - Training Orchestrator → Federated Engine: selects engine
train.py - Federated Engine → UFGraphFR Client Model: trains client
engine.py - Federated Engine → MLP Baseline: trains baseline
mlp.py - Federated Engine → User Graph Utilities: constructs relation graph
engine.py - Federated Engine → Optional Noise: may add noise
README.md - UFGraphFR Client Model → Sequence Transformer: uses attention
mymodel.py - Federated Engine → Ranking Metrics: initializes metrics
engine.py - Training Orchestrator → Evaluation Results: writes results
train.py - Training Orchestrator → Ranking Metrics: evaluates