LinkedIn Connection Recommendation System by: Austyn Herman
Outline Introduction Related Works Current Project Conclusion
Introduction To recommendation systems Collaborative Content Based Recommendations Hybrid Goal To provide meaningful content recommendations
Introduction To Recommendation systems Meaningful Content Recommendation Criteria Type of content being recommended Properties of the network Preferences of the user
Related Works “Recommender System for Location-based Social Networks” by Yiving Cheng, Yangru Fang, and Yongqing Yuan Recommendation Criteria Proximity User Cosine Similarity Friend Check In Results
Related Works “Who Should I Interact With” by Quan Trong, Xiao Chen, and David Frank Recommendation Criteria Page Rank Cosine Similarity Results “Rich get richer”
Related Works “Personalized Recommendation System for Question to Answer on SuperUser” by Geza Kovac, Arpad Kovac, and Shahriyar Pruiskin Recommendation Criteria Content-based filtering Slow start compensation Results Low diversification of questions
Current project Background Methodology Issues, Solutions, and Future Improvements Current project
LinkedIn Friend Recommender System Hybrid Filtering Background
Methodology: Data Collection Data collection methods LinkedIn API Issues Disconnected Graph Solutions Small World Graph Refine API call methods
Methodology: Generating a Recommendation Generating recommendation data Using previous and newly established connections Analysis on recommendation data Cosine Similarity
Issues, Solutions, and Future Improvements Reducing Computation Reducing Referral Selection Pool Slow Start Phase Use of Cosine Similarity Threshold No Reduction on the Selection Pool
Conclusion Recommendation Systems Related Works Current Project
Questions?