We’ve come to expect personalized experiences online—whether it’s Netflix recommending a show or an online retailer suggesting items you might also like to purchase. But how are these suggestions generated? In this course, you’ll learn everything you need to know to create your own recommendation engine. Through hands-on exercises, you’ll get to grips with the two most common systems, collaborative filtering and content-based filtering. Next, you’ll learn how to measure similarities like the Jaccard distance and cosine similarity, and how to evaluate the quality of recommendations on test data using the root mean square error (RMSE). By the end of this course, you’ll have built your very own movie recommendation engine and be able to apply your Python skills to create these systems for any industry.
Introduction to Recommendation EnginesFree
What problems are recommendation engines designed to solve and what data are best suited for them? Discern what insightful recommendations can be made even with limited data, and learn how to create your own recommendations.What are recommendation engines?50 xpRecommendation engines vs. predictions100 xpIdentifying the correct data for recommendation engines50 xpImplicit vs. explicit data100 xpNon-personalized recommendations50 xpIntroduction to non-personalized recommendations100 xpImproved non-personalized recommendations100 xpCombining popularity and reviews100 xpNon-personalized suggestions50 xpFinding all pairs of movies100 xpCounting up the pairs100 xpMaking your first movie recommendations100 xp
Discover how item attributes can be used to make recommendations. Create valuable comparisons between items with both categorical and text data. Generate profiles to recommend new items for users based on their past preferences.Intro to content-based recommendations50 xpWhy use content-based models?50 xpCreating content-based data100 xpUnderstanding the content-based data50 xpMaking content-based recommendations50 xpComparing individual movies with Jaccard similarity100 xpComparing all your movies at once100 xpMaking recommendations based on movie genres100 xpText-based similarities50 xpInstantiate the TF-IDF model100 xpCreating the TF-IDF DataFrame100 xpComparing all your movies with TF-IDF100 xpMaking recommendations with TF-IDF100 xpUser profile recommendations50 xpBuild the user profiles100 xpUser profile based recommendations100 xp
Discover new items to recommend to users by finding others with similar tastes. Learn to make user-based and item-based recommendations—and in what context they should be used. Use k-nearest neighbors models to leverage the wisdom of the crowd and predict how someone might rate an item they haven’t yet encountered.Collaborative filtering50 xpPivoting your data100 xpFinding similar users50 xpChallenges with missing values100 xpCompensating for incomplete data100 xpFinding similarities50 xpUser-based to item-based100 xpSimilar and different movie ratings100 xpFinding similarly liked movies100 xpUsing K-nearest neighbors50 xpStepping through K-nearest neighbors100 xpGetting KNN data in shape100 xpKNN predictions100 xpItem-based or user-based50 xpComparing item-based and user-based models100 xpWhich should you choose?100 xp
Matrix Factorization and Validating Your Predictions
Understand how the sparsity of real-world datasets can impact your recommendations. Leverage the power of matrix factorization to deal with this sparsity. Explore the value of latent features and use them to better understand your data. Finally, put the models you’ve discovered to the test by learning how to validate each of the approaches you’ve learned.Dealing with sparsity50 xpMatrix sparsity100 xpLimited data in your rows100 xpMatrix multiplication50 xpMatrix factorization50 xpIdentifying latent features50 xpInformation loss in factorization100 xpSingular value decomposition (SVD)50 xpNormalize your data100 xpDecomposing your matrix100 xpRecalculating the matrix100 xpMaking recommendations with SVD100 xpValidating your predictions50 xpCalculating RMSE50 xpComparing recommendation methods100 xpWrap up50 xp
Robert O'CallaghanSee More
Director of Data Science, Ordergroove
Rob enables retailers and brands to make themselves indispensable to their customers’ lives by anticipating purchasing needs. Throughout his career, Rob has focused on the analysis, visualization, and modeling of data to produce actionable business improvements for some of the world’s largest organizations. He has successfully designed and implemented multi-million dollar machine learning solutions within several Fortune 500 companies, focusing in particular on bleeding edge unsupervised and supervised learning techniques. He has presented his work, in the U.S. and abroad, to audiences of hundreds at financial services and AI-focused conferences.