class09

# class09 - Ranking Problems 9.520 Class 09, 08 March 2006...

This preview shows pages 1–7. Sign up to view the full content.

Ranking Problems 9.520 Class 09, 08 March 2006 Giorgos Zacharia

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Supervised Ranking Problems Preference Modeling: Given a set of possible product configurations x 1 , x 2 ,…x d predict the most preferred one; predict the rating Information Retrieval: Given a query q, and set of candidate matches x 1 , x 2 ,…x d predict the best answer Information Extraction: Given a set of possible part of speech tagging choices, x 1 , x 2 ,…x d predict the most correct tag boundaries E.g “The_day_they_shot_John_Lennon/WE at the Dogherty_Arts_Center/WE” Multiclass classification: Given a set of possible class labels y 1 , y 2 ,…y d and confindense scores c 1 , c 2 ,…c d , predict the correct label
Types of information available • Preference modeling: – Metric based: • User rated configuration x with y=U (x) i i i – Choice based: • Given choices x 1 , x 2 ,…x d , the user chose x f – Prior information about the features: • Cheaper is better • Faster is better •e t c

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Types of information available • Information Retrieval: – Metric based: • Users clicked on link x i with a frequency y=U (x) i i – Choice based: • Given choices x 1 , x 2 ,…x d , the user clicked on x f – Prior information about the features: • Keyword matches (the more the better) • Unsupervised similarity scores (TFIDF) •e t c
Types of information available • Information Extraction: Choice based: • Given tagging choices x 1 , x 2 ,…x d , the hand labeling chose x f Prior information about the features: Unsupervised scores • Multiclass: Choice based: Given vectors the confidence scores c 1 , c 2 ,…c d for class labels 1,2,…d the correct label was y f.. . The confidence scores may be coming from set of weak classifiers, and/or OVA comparisons. Prior information about the features: The higher the confidence score the more likely to represent the correct label.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
(Semi-)Unsupervised Ranking Problems Learn relationships of the form: Class A is closer to B, than it is to C We are given a set of l labeled comparisons for a user, and a set of u seemingly-unrelated comparisons from other users.
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 11/11/2011 for the course BIO 9.07 taught by Professor Ruthrosenholtz during the Spring '04 term at MIT.

### Page1 / 27

class09 - Ranking Problems 9.520 Class 09, 08 March 2006...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online