USER-BASED COLLABORATIVE FILTERING
A large part of recommender systems is deciphering whether a user will be receptive to a recommended item. Singular-value decomposition, as explored in the previous chapter, is a method that can be used to predict user preferences in the case of a sparse dataset. For situations where data is not sparse, we can use other algorithms including logistic regression and Naive Bayes.
The Naive Bayes Classifier—a classification method based on Bayes’ theorem—is often used in data science for predicting discrete classes such as spam filtering, A/B testing, sentiment analysis, text classification, and recommender systems.
Bayes’ Theorem
Devised in the middle of the 18th Century, Bayes' theorem is used for making inferences based on existing information (conditional probability) and is a central pillar of classical statistics. The formula for Bayes’ theorem is shown here:
In this formula...