Building a Graph Machine Learning App with Streamlit and Graph Convolutional Networks

## Introduction Graph-structured data is everywhere in the modern world. Social networks, recommendation systems, citation networks, transportation systems, and knowledge graphs all represent information as nodes connected by relationships. Traditional machine learning techniques often struggle to capture these complex relationships because they are designed primarily for tabular or sequential data. Graph Machine Learning (Graph ML) has emerged as a powerful field that enables deep learning models to learn directly from graph structures. One of the most influential architectures in this domain is the Graph Convolutional Network (GCN), which extends the concept of convolution from images to graphs. The **Graph Machine Learning App** is a Streamlit-based application that demonstrates node classification using a Graph Convolutional Network trained on the widely used Cora citation dataset. The application provides an interactive interface where users can train the model, monitor training progress, and evaluate classification accuracy in real time. --- ## Problem Statement Many real-world datasets contain interconnected entities rather than independent records. For example: * Research papers cite other papers. * Users interact with other users in social networks. * Products are linked through purchasing behavior. * Web pages connect through hyperlinks. Traditional machine learning models typically ignore these relationships and focus only on individual features. As a result, they may fail to capture important structural information. The challenge is to build models that learn from both: * Node features * Graph connectivity Graph Neural Networks address this challenge by propagating information across neighboring nodes and learning meaningful graph representations. This application demonstrates how Graph Convolutional Networks can be used to classify nodes within a citation network. --- ## Features The Graph Machine Learning App provides several important features: ### Cora Dataset Integration The application automatically loads the Cora citation network dataset using PyTorch Geometric. ### Graph Convolutional Network (GCN) A two-layer GCN architecture is implemented for node classification. ### Interactive Training Users can start model training through a simple button click. ### Real-Time Progress Tracking A progress bar displays training status across epochs. ### Loss Monitoring Training loss is updated dynamically during model training. ### Model Evaluation The application calculates and displays node classification accuracy after training. ### Streamlit-Based Interface A clean and user-friendly interface enables experimentation without requiring command-line interaction. --- ## Technologies Used The project combines modern graph learning and web technologies. | Technology | Purpose | | ---------------------------------- | ---------------------------------- | | Python | Core programming language | | Streamlit | Interactive web application | | PyTorch | Deep learning framework | | PyTorch Geometric | Graph machine learning library | | Graph Convolutional Networks (GCN) | Graph neural network architecture | | Cora Dataset | Benchmark citation network dataset | Together, these technologies create a practical environment for experimenting with graph-based deep learning. --- ## How It Works The application uses the Cora citation dataset, a benchmark graph dataset commonly used in graph machine learning research. The dataset consists of: * Research papers represented as nodes * Citation relationships represented as edges * Paper features represented as node attributes * Research categories represented as labels The model architecture contains two Graph Convolutional layers: ### Layer 1 ```python GCNConv(dataset.num_features, 16) ``` This layer aggregates information from neighboring nodes and transforms it into a hidden representation. ### Layer 2 ```python GCNConv(16, dataset.num_classes) ``` This layer generates classification scores for each node. During training: 1. Node features are propagated through the graph. 2. The model predicts node classes. 3. Cross-entropy loss is calculated. 4. Backpropagation updates model weights. 5. Training repeats for 200 epochs. After training, the model is evaluated on the test nodes and accuracy is calculated. --- ## Application Workflow ### Step 1: Load Dataset The application loads the Cora citation graph using PyTorch Geometric. ### Step 2: Initialize Model A Graph Convolutional Network is created. ### Step 3: Start Training The user clicks: ```text 🚀 Train Model ``` ### Step 4: Training Progress The application displays: * Current epoch * Training loss * Progress bar ### Step 5: Model Evaluation After training, predictions are generated for test nodes. ### Step 6: Display Results The final accuracy score is shown in the Streamlit interface. --- ## Example Input The application uses the built-in Cora dataset. Dataset Characteristics: ```text Nodes (Research Papers): 2708 Edges (Citations): 10556 Features per Node: 1433 Classes: 7 ``` The user simply clicks: ```text 🚀 Train Model ``` No manual data upload is required. --- ## Example Output During training: ```text Epoch 50/200 | Loss: 0.8421 Epoch 100/200 | Loss: 0.5328 Epoch 150/200 | Loss: 0.3417 Epoch 200/200 | Loss: 0.2154 ``` Training completion: ```text ✅ Training Complete! ``` Results section: ```text 📈 Results Accuracy: 0.8120 ``` Actual accuracy may vary slightly between training runs due to random initialization. --- ## Use Cases Graph Machine Learning has applications across many industries. ### Citation Analysis Classify research papers based on citation relationships. ### Social Networks Predict user interests and community memberships. ### Recommendation Systems Recommend products, movies, or content using graph structures. ### Fraud Detection Identify suspicious transactions in financial networks. ### Knowledge Graphs Improve search and question-answering systems. ### Healthcare Analyze patient and disease relationship networks. ### Cybersecurity Detect malicious behavior in network traffic graphs. --- ## Future Improvements Several enhancements can further improve the application. ### Graph Visualization Display interactive graph structures within Streamlit. ### Multiple Datasets Support additional graph datasets such as: * CiteSeer * PubMed * OGB datasets ### Advanced Architectures Integrate more sophisticated graph neural networks such as: * GraphSAGE * GAT (Graph Attention Networks) * GIN (Graph Isomorphism Networks) ### Hyperparameter Tuning Allow users to customize: * Learning rate * Number of layers * Hidden dimensions * Epoch count ### Model Comparison Dashboard Compare multiple graph models side-by-side. ### Performance Metrics Add precision, recall, and F1-score visualizations. --- ## Conclusion The Graph Machine Learning App demonstrates how Graph Neural Networks can be applied to graph-structured data using a simple and interactive Streamlit interface. By leveraging the Cora citation dataset and a Graph Convolutional Network architecture, the application provides a practical introduction to node classification and graph-based deep learning. The project highlights the growing importance of Graph Machine Learning in modern AI applications, where relationships between entities are often as important as the entities themselves. Through Streamlit and PyTorch Geometric, complex graph learning concepts become accessible to students, researchers, and developers interested in exploring the rapidly evolving field of Graph AI.