Facebook Friend Recommendation using Graph Mining

Project Overview

This graph mining case study addresses the link prediction challenge in social networks. By representing user relationships as nodes and edges in a directed graph, the model analyzes existing network topologies to predict missing links and suggest potential friend recommendations.

Key Features & Objectives

• Link Prediction: Analyzes connectivity structures to identify high-probability future connections.
• Graph Telemetry: Maps network metrics (shortest path, connected components) to represent profile relationships.
• Matrix Representation: Encodes network adjacency features into mathematical vectors to train classifiers.

Technical Stack & Technologies

• Language: Python
• Graph Framework: NetworkX
• Machine Learning: Scikit-learn, XGBoost
• Data Structures: SciPy Sparse Matrices, Pandas, NumPy

Core Techniques & Graph Algorithms

• Similarity Metrics Extraction: Engineered graph-theoretic features, including Jaccard Distance, Cosine Distance, PageRank scores, HITS hub/authority metrics, and the Adamic-Adar Index.
• SVD Matrix Decomposition: Decomposed the graph’s adjacency matrix using Singular Value Decomposition (SVD) to represent low-dimensional structural connection attributes.
• Gradient-Boosted Classifier: Trained an XGBoost model utilizing the extracted graph features to classify if a directed link between two user nodes is likely to exist.