Recommender Systems Tutorial (Part 2) -- Offline Components

Offline Components:
Collaborative Filtering in Cold-start Situations

Problem

Algorithm selects Item j with item features xj
(keywords, content categories, ...)

User i visits (i, j) : response yij
with (explicit rating, implicit click/no-click)
user features xi
(demographics,
browse history,
Predict the unobserved entries based on
search history, …) features and the observed entries

Deepak Agarwal & Bee-Chung Chen @ ICML’11 2

Model Choices

• Feature-based (or content-based) approach
– Use features to predict response
• (regression, Bayes Net, mixture models, …)
– Limitation: need predictive features
• Bias often high, does not capture signals at granular levels

• Collaborative filtering (CF aka Memory based)
– Make recommendation based on past user-item interaction
• User-user, item-item, matrix factorization, …
• See [Adomavicius & Tuzhilin, TKDE, 2005], [Konstan, SIGMOD’08
Tutorial], etc.
– Better performance for old users and old items
– Does not naturally handle new users and new items (cold-start)


Collaborative Filtering (Memory based methods)

User-User Similarity

Item-Item similarities, incorporating both
Estimating Similarities
• Pearson’s correlation
• Optimization based (Koren et al)


How to Deal with the Cold-Start Problem

• Heuristic-based approaches
– Linear combination of regression and CF models

– Filterbot
• Add user features as psuedo users and do collaborative filtering
- Hybrid approaches
- Use content based to fill up entries, then use CF

• Matrix Factorization
– Good performance on Netflix (Koren, 2009)
• Model-based approaches
– Bilinear random-effects model (probabilistic matrix factorization)
• Good on Netflix data [Ruslan et al ICML, 2009]
– Add feature-based regression to matrix factorization
• (Agarwal and Chen, 2009)
– Add topic discovery (from textual items) to matrix factorization
• (Agarwal and Chen, 2009; Chun and Blei, 2011)

Per-item regression models

• When tracking users by cookies, distribution of visit patters
could get extremely skewed
– Majority of cookies have 1-2 visits

• Per item models (regression) based on user covariates
attractive in such cases


Several per-item regressions: Multi-task learning

• Agarwal,Chen and Elango, KDD, 2010

•Low dimension
•(5-10),

•B estimated
•retrospective data

Affinity to
old items


Per-user, per-item models
via bilinear random-effects model

Motivation

• Data measuring k-way interactions pervasive
– Consider k = 2 for all our discussions
• E.g. User-Movie, User-content, User-Publisher-Ads,….
– Power law on both user and item degrees

• Classical Techniques
– Approximate matrix through a singular value decomposition (SVD)
• After adjusting for marginal effects (user pop, movie pop,..)
– Does not work
• Matrix highly incomplete, severe over-fitting
– Key issue
• Regularization of eigenvectors (factors) to avoid overfitting


Early work on complete matrices

• Tukey’s 1-df model (1956)
– Rank 1 approximation of small nearly complete matrix

• Criss-cross regression (Gabriel, 1978)
• Incomplete matrices: Psychometrics (1-factor model only;
small data sets; 1960s)
• Modern day recommender problems
– Highly incomplete, large, noisy.


Latent Factor Models
•u •s
p
o •v
r
t •Affinity = u’v
•“newsy”
y

•item
•“sporty” •s

•z

•“newsy” •Affinity = s’z


Factorization – Brief Overview

• Latent user factors: • Latent movie factors:
(αi , ui=(ui1,…,uin)) (βj , vj=(v j1,….,v jn))
Interaction

E ( yij ) = µ + α i + β j + ui′Bv j

• (Nn + Mm) will overfit for moderate
values of n,m
parameters

Regularization
• Key technical issue:


Latent Factor Models: Different Aspects

• Matrix Factorization
– Factors in Euclidean space
– Factors on the simplex

• Incorporating features and ratings simultaneously
• Online updates


Maximum Margin Matrix Factorization (MMMF)

• Complete matrix by minimizing loss (hinge,squared-error)
on observed entries subject to constraints on trace norm
– Srebro, Rennie, Jakkola (NIPS 2004)
• Convex, Semi-definite programming (expensive, not scalable)

• Fast MMMF (Rennie & Srebro, ICML, 2005)
– Constrain the Frobenious norm of left and right eigenvector
matrices, not convex but becomes scalable.

• Other variation: Ensemble MMMF (DeCoste, ICML2005)
– Ensembles of partially trained MMMF (some improvements)


Matrix Factorization for Netflix prize data

• Minimize the objective function

∑ (r − ui v j ) + λ ( ∑ ui + ∑ v j )
T 2 2 2
ij
ij∈obs i j

• Simon Funk: Stochastic Gradient Descent

• Koren et al (KDD 2007): Alternate Least Squares
– They move to SGD later in the competition


Probabilistic Matrix Factorization
(Ruslan & Minh, 2008, NIPS)
au av •Optimization is through Gradient
Descent (Iterated conditional modes)

ui vj
•Other variations like constraining the
mean through sigmoid, using “who-rated-
whom”

•Combining with Boltzmann Machines
also improved performance
rij
rij ~ N (uT v j , σ 2 )
i

u i ~ MVN (0, au I )
σ 2
v j ~ MVN (0, av I )


Bayesian Probabilistic Matrix Factorization
(Ruslan and Minh, ICML 2008)

• Fully Bayesian treatment using an MCMC approach
• Significant improvement

• Interpretation as a fully Bayesian hierarchical model
shows why that is the case
– Failing to incorporate uncertainty leads to bias in estimates
– Multi-modal posterior, MCMC helps in converging to a better one

MCEM also more resistant to over-fitting
•r
•Var-comp: au


Non-parametric Bayesian matrix completion
(Zhou et al, SAM, 2010)

• Specify rank probabilistically (automatic rank selection)
r
yij ~ N (∑ z k uik v jk , σ ) 2

k =1

z k ~ Ber (π k )
π k ~ Beta(a / r , b(r − 1) / r )
z k ~ Ber (1, a /( a + b( r −1)))
E ( # Factors) = ra /( a + b( r −1))


How to incorporate features:
Deal with both warm start and cold-start

• Models to predict ratings for new pairs
– Warm-start: (user, movie) present in the training data with
large sample size
– Cold-start: At least one of (user, movie) new or has small
sample size
• Rough definition, warm-start/cold-start is a continuum.

• Challenges
– Highly incomplete (user, movie) matrix
– Heavy tailed degree distributions for users/movies
• Large fraction of ratings from small fraction of users/movies
– Handling both warm-start and cold-start effectively in the
presence of predictive features

Possible approaches

• Large scale regression based on covariates
– Does not provide good estimates for heavy users/movies
– Large number of predictors to estimate interactions

• Collaborative filtering
– Neighborhood based
– Factorization
• Good for warm-start; cold-start dealt with separately

• Single model that handles cold-start and warm-start
– Heavy users/movies → User/movie specific model
– Light users/movies → fallback on regression model
– Smooth fallback mechanism for good performance


Add Feature-based Regression into
Matrix Factorization

RLFM: Regression-based Latent Factor Model

Regression-based Factorization Model (RLFM)

• Main idea: Flexible prior, predict factors through
regressions
• Seamlessly handles cold-start and warm-start
• Modified state equation to incorporate covariates


RLFM: Model

• Rating: yij ~ N ( µij , σ 2 ) Gaussian Model
user i yij ~ Bernoulli ( µij ) Logistic Model (for binary rating)
gives yij ~ Poisson( µij N ij ) Poisson Model (for counts)
item j

t ( µij ) = xij b + α i + β j+ui t v j
t

• Bias of user i: α i = g 0 xi + ε iα , ε iα ~ N (0, σ α )
t 2

• Popularity of item j: β j = d 0 x j + ε β , ε β ~ N (0, σ β )
t
j j
2

• Factors of user i: ui = Gxi + ε iu , ε iu ~ N (0,σ u I )
2

• Factors of item j: vi = Dx j + ε iv , ε iv ~ N (0,σ v I )
2

Could use other classes of regression models


Graphical representation of the model


Advantages of RLFM

• Better regularization of factors
– Covariates “shrink” towards a better centroid

• Cold-start: Fallback regression model (FeatureOnly)


RLFM: Illustration of Shrinkage

Plot the first factor
value for each user
(fitted using Yahoo! FP
data)


Induced correlations among observations

Hierarchical random-effects model

Marginal distribution obtained by
integrating out random effects


Closer look at induced marginal correlations


Model fitting: EM for our class of models


The parameters for RLFM

• Latent parameters

Δ = ({αi }, {β j }, {ui }, {v j })

• Hyper-parameters

Θ = (b, G, D, A u = a u I, A v = a v I)


Computing the mode

Minimized


The EM algorithm


Computing the E-step

• Often hard to compute in closed form
• Stochastic EM (Markov Chain EM; MCEM)
– Compute expectation by drawing samples from

– Effective for multi-modal posteriors but more expensive
• Iterated Conditional Modes algorithm (ICM)
– Faster but biased hyper-parameter estimates


Monte Carlo E-step

• Through a vanilla Gibbs sampler (conditionals closed form)

• Other conditionals also Gaussian and closed form
• Conditionals of users (movies) sampled simultaneously
• Small number of samples in early iterations, large numbers
in later iterations

M-step (Why MCEM is better than ICM)

• Update G, optimize

• Update Au=au I

Ignored by ICM, underestimates factor variability
Factors over-shrunk, posterior not explored well


Experiment 1: Better regularization

• MovieLens-100K, avg RMSE using pre-specified splits
• ZeroMean, RLFM and FeatureOnly (no cold-start issues)
• Covariates:
– Users : age, gender, zipcode (1st digit only)
– Movies: genres


Experiment 2: Better handling of Cold-start

• MovieLens-1M; EachMovie
• Training-test split based on timestamp
• Same covariates as in Experiment 1.


Experiment 4: Predicting click-rate on articles

• Goal: Predict click-rate on articles for a user on F1 position
• Article lifetimes short, dynamic updates important

• User covariates:
– Age, Gender, Geo, Browse behavior

• Article covariates
– Content Category, keywords

• 2M ratings, 30K users, 4.5 K articles


Results on Y! FP data


Some other related approaches

• Stern, Herbrich and Graepel, WWW, 2009
– Similar to RLFM, different parametrization and expectation
propagation used to fit the models

• Porteus, Asuncion and Welling, AAAI, 2011
– Non-parametric approach using a Dirichlet process

• Agarwal, Zhang and Mazumdar, Annals of Applied
Statistics, 2011
– Regression + random effects per user regularized through a
Graphical Lasso


Add Topic Discovery into
Matrix Factorization

fLDA: Matrix Factorization through Latent Dirichlet Allocation

fLDA: Introduction

• Model the rating yij that user i gives to item j as the user’s
affinity to the topics that the item has
User i ’s affinity to topic k

yij = ... + ∑k sik z jk
Pr(item j has topic k) estimated by averaging
the LDA topic of each word in item j
Old items: zjk’s are Item latent factors learnt from data with the LDA prior
New items: zjk’s are predicted based on the bag of words in the items

– Unlike regular unsupervised LDA topic modeling, here the LDA
topics are learnt in a supervised manner based on past rating data
– fLDA can be thought of as a “multi-task learning” version of the
supervised LDA model [Blei’07] for cold-start recommendation

LDA Topic Modeling (1)

• LDA is effective for unsupervised topic discovery [Blei’03]
– It models the generating process of a corpus of items (articles)
– For each topic k, draw a word distribution Φk = [Φk1, …, ΦkW] ~ Dir(η)
– For each item j, draw a topic distribution θj = [θj1, …, θjK] ~ Dir(λ)
– For each word, say the nth word, in item j,
• Draw a topic zjn for that word from θj = [θj1, …, θjK]
• Draw a word wjn from Φk = [Φk1, …, ΦkW] with topic k = zjn

Item j
Topic distribution: [θj1, …, θjK] Φ11, …, Φ1W Topic 1
Assume zjn = topic k
…
Per-word topic: zj1, …, zjn, … Φk1, …, ΦkW Topic k
…
ΦK1, …, ΦKW Topic K
Words: wj1, …, wjn, …
Observed

LDA Topic Modeling (2)

• Model training:
– Estimate the prior parameters and the posterior topic×word
distribution Φ based on a training corpus of items
– EM + Gibbs sampling is a popular method
• Inference for new items
– Compute the item topic distribution based on the prior parameters
and Φ estimated in the training phase
• Supervised LDA [Blei’07]
– Predict a target value for each item based on supervised LDA
topics
Regression weight for topic k One regression per user

yj = ∑ k
sk z jk vs. yij = ... + ∑k sik z jk
Same set of topics across different regressions
Target value of item j
Pr(item j has topic k) estimated by averaging
the topic of each word in item j

fLDA: Model

• Rating: yij ~ N ( µij , σ 2 ) Gaussian Model
user i yij ~ Bernoulli ( µij ) Logistic Model (for binary rating)
gives yij ~ Poisson( µij N ij ) Poisson Model (for counts)
item j

t ( µij ) = xij b + α i + β j + ∑ k sik z jk
t

• Bias of user i: α i = g 0 xi + ε iα , ε iα ~ N (0, σ α )
t 2

• Popularity of item j: β j = d 0 x j + ε β , ε β ~ N (0, σ β )
t
j j
2

• Topic affinity of user i: si = Hxi + ε is , ε is ~ N (0, σ s2 I )
• Pr(item j has topic k): z jk = ∑ n 1( z jn = k ) /( # words in item j )
The LDA topic of the nth word in item j
• Observed words: w jn ~ LDA(λ ,η , z jn )
The nth word in item j

Model Fitting

• Given:
– Features X = {xi, xj, xij}
– Observed ratings y = {yij} and words w = {wjn}

• Estimate:
– Parameters: Θ = [b, g0, d0, H, σ2, aα, aβ, As, λ, η]
• Regression weights and prior parameters
– Latent factors: ∆ = {αi, βj, si} and z = {zjn}
• User factors, item factors and per-word topic assignment
• Empirical Bayes approach:
– Maximum likelihood estimate of the parameters:
ˆ
∫
Θ = arg max Pr[ y, w | Θ] = arg max Pr[ y, w, ∆, z | Θ] d∆dz
Θ Θ
– The posterior distribution of the factors:

ˆ
Pr[∆, z | y, Θ]

The EM Algorithm

• Iterate through the E and M steps until convergence
ˆ
– Let Θ( n ) be the current estimate
– E-step: Compute
f ( Θ) = E ˆ [log Pr( y , w, ∆, z
( ∆ , z | y , w, Θ n )
| Θ)]
• The expectation is not in closed form
• We draw Gibbs samples and compute the Monte Carlo mean

ˆ
Θ( n +1) = arg max f (Θ)
– M-step: Find
Θ

• It consists of solving a number of regression and optimization
problems


Supervised Topic Assignment

The topic of the nth word in item j
Probability of observing yij
Pr( z jn = k | Rest ) given the model

Z kl jn + η
¬
∝ ¬jn
Z k + Wη
jk ( )
Z ¬jn + λ ⋅ ∏i rated j f ( yij | z jn = k )

Same as unsupervised LDA Likelihood of observed ratings
by users who rated item j when
zjn is set to topic k


fLDA: Experimental Results (Movie)

• Task: Predict the rating that a user would give a movie
• Training/test split:
Model Test RMSE
– Sort observations by time RLFM 0.9363
– First 75% → Training data fLDA 0.9381
Factor-Only 0.9422
– Last 25% → Test data FilterBot 0.9517
unsup-LDA 0.9520
• Item warm-start scenario MostPopular 0.9726
– Only 2% new items in test data Feature-Only 1.0906
Constant 1.1190

fLDA is as strong as the best method
It does not reduce the performance in warm-start scenarios


fLDA: Experimental Results (Yahoo! Buzz)

• Task: Predict whether a user would buzz-up an article
• Severe item cold-start
– All items are new in test data

fLDA significantly
outperforms other
models

Data Statistics
1.2M observations
4K users
10K articles


Experimental Results: Buzzing Topics

3/4 topics are interpretable; 1/2 are similar to unsupervised topics
Top Terms (after stemming) Topic
bush, tortur, interrog, terror, administr, CIA, offici, CIA interrogation
suspect, releas, investig, georg, memo, al
mexico, flu, pirat, swine, drug, ship, somali, border, Swine flu
mexican, hostag, offici, somalia, captain
NFL, player, team, suleman, game, nadya, star, high, NFL games
octuplet, nadya_suleman, michael, week
court, gai, marriag, suprem, right, judg, rule, sex, pope, Gay marriage
supreme_court, appeal, ban, legal, allow
palin, republican, parti, obama, limbaugh, sarah, rush, Sarah Palin
gop, presid, sarah_palin, sai, gov, alaska
idol, american, night, star, look, michel, win, dress, American idol
susan, danc, judg, boyl, michelle_obama
economi, recess, job, percent, econom, bank, expect, Recession
rate, jobless, year, unemploy, month
north, korea, china, north_korea, launch, nuclear, North Korea issues
rocket, missil, south, said, russia


fLDA Summary

• fLDA is a useful model for cold-start item recommendation
• It also provides interpretable recommendations for users
– User’s preference to interpretable LDA topics
• Future directions:
– Investigate Gibbs sampling chains and the convergence properties of the
EM algorithm
– Apply fLDA to other multi-task prediction problems
• fLDA can be used as a tool to generate supervised features
(topics) from text data


Summary

• Regularizing factors through covariates effective

• Regression based factor model that regularizes better and
deals with both cold-start and warm-start in a single
framework in a seamless way looks attractive

• Fitting method scalable; Gibbs sampling for users and
movies can be done in parallel. Regressions in M-step can
be done with any off-the-shelf scalable linear regression
routine
• Distributed computing on Hadoop: Multiple models and
average across partitions

Hierarchical smoothing
Advertising application, Agarwal et al. KDD 10

• Product of states for each node pair
Advertiser
pubclass
Conv-id

campaign
Pub-id
z Ad-id

• Spike and Slab prior ( Sz, Ez, λz)

– Known to encourage parsimonious solutions
• Several cell states have no corrections
– Not used before for multi-hierarchy models, only in regression
– We choose P = .5 (and choose “a” by cross-validation)
• a – psuedo number of successes


Recommender Systems Tutorial (Part 2) -- Offline Components

More Related Content

What's hot (18)

Similar to Recommender Systems Tutorial (Part 2) -- Offline Components (20)

Recently uploaded (20)

Recommender Systems Tutorial (Part 2) -- Offline Components

Editor's Notes