OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe...

21
OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren , Joe Sill Recsys’11 best paper award

Transcript of OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe...

Page 1: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

OrdRec: An Ordinal Model for Predicting Personalized Item

Rating Distributions

Yehuda Koren , Joe SillRecsys’11 best paper award

Page 2: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Outline

Motivations The OrdRec Model MultiNomial Factor Model Experiment

Page 3: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Motivations

Numerical v.s. Ordinal

Page 4: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Motivations

A comparative ranking of products No direct interpretation in terms of

numerical values Numerical may not reflect user

intention well User bias

Page 5: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Motivations

OrdRec Model Motivated by above discussion and

inspired by the ordinal logistic regression model by McCullagh

Ability to output a full probability distribution of the scores

Ability to associated with confidence levels

Page 6: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

The OrdRec Model

)|)(|()(

2/1

uRjju

Tiuiui xuRpqbby

121 sttt

)exp(1 rrr tt 2,,1 Sr

Page 7: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

The OrdRec Model

)1,( uiui yNz

)()|( 1 ruirui tztPrrP

)()()|( uirruiui yttzPrrP

Page 8: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

The OrdRec Model

)|1()|()|( rrPrrPrrP uiuiui

),,(

)|(log()(riu

ui rrPRL

Page 9: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Ranking items for a user

OrdRec predicts a full probability distribution over ratings Much richer output

Rank items given predicted rating distributions Computing Statistics like mean no

longer plausible Cast the problem as a learning-to-rank

task

Page 10: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Ranking items for a user

)},(,|),,{( jiPrrjiT uujui

),,(

)0()|(ji

TwIwTF

Page 11: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

A multinomial Factor Model(MultiMF)

A multinomial distribution over categorical scores Constructed baseline model for

comparing with OrdRec For each score r:

Same as OrdRec, log likelihood of training data is maximized

ri

T

uRjju

ru

ri

rui qxuRpbbz )|)(|(

)(

2/1

S

r

rui

rui

ui

z

zrrP

1'

' )exp(

)exp()|(

),,(

)|(log()(riu

ui rrPRL

Page 12: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Experiments

Data set Netflix Two Yahoo! Music Data set

Page 13: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Evaluation Metrics

uuduc

uuc

nn

nFCP

)(

Page 14: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Results

Page 15: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Result Analysis

OrdRec as leader on Nexflix for both RMSE and FCP. Better model ordinal semantics of user ratings

SVD++ performs best in terms of RMSE The only methods trained to minimize RMSE

RMSE values on Y!Music much greater than Netflix while FCP values changes little RMSE more sensitive to rating scales than

FCP (Y!Music 10 scales, Netflix 5 scales)

Page 16: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Result Analysis

OrdRec consistently outperforms the rest in terms of FCP Indicate it better ranking items for a user:

reflect the benefit of better modeling the semantics of user feedback

Training time comparison

Page 17: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Recommendation Confidence Estimation

Formulate confidence estimation as a binary classification problem Predict whether the model’s predicted

rating is within one rating level of the true rating

Predicted values : expected value of the predicted rating distribution

Using logistic regression to predict Random 2/3 of Test data as training , the

rest as test

Page 18: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Result

Page 19: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Conclusions

Taking user feedback as ordinal relaxes the numerical view Can deal with all usual feedbacks, such

as thumbs-up/down, like-votes, stars, numerical scores, or A-F grades

Without assuming categorical feedback Also applied even feedback is actual

numerical: It allow expresses distinct internal scales for their qualitative ratings

Page 20: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Conclusions

OrdRec employs a point-wise approach to ordinal modeling Training time is linearly with data set

size OrdRec outputs a full probability

distribution of scores Provides richer expressive power Helpful in estimating the confidence

level

Page 21: OrdRec: An Ordinal Model for Predicting Personalized Item Rating Distributions Yehuda Koren, Joe Sill Recsys’11 best paper award.

Thank you Q&A