• Save
IJCAI Workshop Presentation
Upcoming SlideShare
Loading in...5
×
 

IJCAI Workshop Presentation

on

  • 896 views

 

Statistics

Views

Total Views
896
Views on SlideShare
878
Embed Views
18

Actions

Likes
0
Downloads
0
Comments
0

3 Embeds 18

http://www0.cs.ucl.ac.uk 14
http://www.slideshare.net 2
http://www.cs.ucl.ac.uk 2

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    IJCAI Workshop Presentation IJCAI Workshop Presentation Presentation Transcript

    • neal lathia
    • finishing PhD @ UCL (London) S. Hailes & L. Capra intern @ Telefonica I+D (Barcelona) X. Amatriain & J. M. Pujol
    • collaborative filtering
    • statistics
    • statistics user modeling
    • what does the how do people data show? make decisions?
    • similarity trust user modeling reputation
    • like- minded? similarity friends? trust user modeling experts? reputation
    • SIGIR '09
    • 1. Get “expert” data 2. Compare experts and Netflix “users” SIGIR '09 3. Recommend to users based on experts 4. Evaluate recommendations
    • experts? Accuracy Top-N Precision User Study
    • experts? Accuracy neighbors? Top-N Precision User Study enthusiasts?
    • neighbors? experts? user enthusiasts?
    • given: a simple, un-tuned, kNN predictor and multiple information sources
    • a problem: users are subjective, accuracy varies with source
    • a problem: users are subjective, accuracy varies with source
    • a promise: optimal classification of users to best source produces incredibly accurate predictions
    • a promise: optimal classification of users to best source produces incredibly accurate predictions
    • a question: how to classify users to source set?
    • preliminary attempts: (supervised/unsupervised) kNN-voting, similarity-based, best-fit, decision trees, SVD, linear combinations, ...unsuccessful
    • preliminary attempts: learning on the features of user profiles (mean, sd, what was rated..) ...unsuccessful
    • metrics: is the overall RMSE improving? is the precision/recall of the classification improving?
    • lessons: (1) the web is a goldmine of ratings – waiting to be harvested (2) recommender systems need to model how people make decisions (3) accuracy is possible without tuning
    • lessons: (2) recommender systems need to model how people make decisions (3) accuracy is possible without tuning
    • lessons: (3) accuracy is possible without tuning: ...from rating prediction to user classification ...from hybrid predictors to hybrid datasets
    • thanks n.lathia@cs.ucl.ac.uk