5. Academia vs Industry
• Academia:
• Beautiful math, complex models
• Small datasets, evaluation on offline metrics (RMSE,
Precision@N)
• Industry:
• Simple models - MostPopular, SVD
• «Big Data», product metrics: CTR, Retention
6. Academia
• Importance of Replicable Evaluation
• More connections with Industry (data and tasks)
• Fighting for 10^-2 improvements in RMSE
7. RecSys in Social Networks
• Friendship (trustdistrust) connections implies you
have similar tastes
• A Probabilistic Model for Using Social Networks in Personalized Item Recommendation
• Overlapping Community Regularization for Rating Prediction in Social Recommender
Systems
8. Distinguished Papers
• Applying Differential Privacy to Matrix Factorization
• Gaussian Ranking by Matrix Factorization - Netflix
• Context-Aware Event Recommendation in Event-
based Social Networks
• Different examples of features for RankML
9. Interesting
• It Takes Two to Tango: an Exploration of Domain Pairs for Cross-Domain
Collaborative Filtering
• CrossDomain Recommendations - use different datasources (books,
music, ..) to obtain «better» latent factors
• Random Walks for very Diverse Recommendations from «Long Tail Items»
• Blockbusters and Wallflowers: Accurate, Diverse, and Scalable
Recommendations with Random Walks
• Prediction of Online-CTR metric based on Many Offline Metrics
• Predicting Online Performance of News Recommender Systems
Through Richer Evaluation Metrics
10. Short Papers
• Music Playlist Generation: Word2Vec, 2-level HMM
• Factorization Machines on some dataset
• ColdStart Problem tricks
• RecSys based on Streaming Data
• Demonstrations: health-aware food
recommendations
11. Industry
• Simple Models on large datasets beats Complex Models on smaller dataset
• Results on offline metrics can differ from results on product metrics => AB testing
• UX design is even more important than models
• Models:
• Logistic Regression, Most Popular, SVD, Word2Vec
• Clustering
• Metrics:
• CTR, retention, «stability of coming back»
• Technologies:
• Casandra, ElasticSearch
• Hadoop, Spark, Kafka
12. Industry Sessions
• LinkedIn
• Architecture: Kafka, Spark
• What Data Scientist do on the job?
• Booking.com
• Distinctive features of their recommendation system
• UX designers works closely with Data Scientists
• Netflix - «Gaussian Ranking by Matrix Factorization» - one of the Distinguished Papers
• Gravity - History of growth
• IBM Research - «Assessing Expertise in the Enterprise»
• No «Silver Bullet» in recommendations systems, domain knowledge
• OpenTable - Restaurant Recommendations
• Cool ways to combine several data sources for recommendations
• Amazon, Pandora, Spotify, ….