0. Introduction
* Define the problems of classical filtering (e.g., news filtering,
spam filtering) and collaborative filtering (e.g., recommendation
systems).
* Compare and contrast: predictions based upon features (spam
filtering) vs. predictions based upon similar preferences
(recommendation systems).
* Look at real-world examples: SpamAssassin as used by CCIS;
collaborative filtering at Amazon.
1. Filtering, e.g. spam.
* Features and feature selection.
* SpamAssassin.
* Linear classifiers in general, and SpamAssassin as
a linear classifier.
* Positively and negatively correlated features.
* Short introduction to training linear classifiers:
separable data vs. non-separable data.
* Evaluation: (1) simple error rates, (2) false positives
vs. false negatives and the idea behind "utility functions"
(though I never mentioned that phrase).
2. Collaborative filtering, e.g. recommendation systems.
* Weighted combination of recommendations from
"like minded" individuals.
* How to assess "like mindedness?" Linear correlation
coefficients.
* Positive correlation vs. negative correlation vs.
no correlation.
Go to ISU535 home page. or RPF's Teaching Gateway or homepage