information filtering si650 : information retrieva l
DESCRIPTION
Information Filtering SI650 : Information Retrieva l. Winter 2010 School of Information University of Michigan. Many slides are from Prof. ChengXiang Zhai’s lecture. Lecture Plan. Filtering vs . Retrieval Content-based filtering (adaptive filtering) - PowerPoint PPT PresentationTRANSCRIPT
2010 © University of Michigan
Information FilteringSI650: Information Retrieval
Winter 2010
School of InformationUniversity of Michigan
Many slides are from Prof. ChengXiang Zhai’s lecture
1
2010 © University of Michigan
Lecture Plan• Filtering vs. Retrieval• Content-based filtering (adaptive filtering)• Collaborative filtering (recommender systems)
2
2010 © University of Michigan
Short vs. Long Term Info Need• Short-term information need (Ad hoc retrieval)
– “Temporary need”, e.g., info about used cars– Information source is relatively static – User “pulls” information– Application example: library search, Web search
• Long-term information need (Filtering)– “Stable need”, e.g., new data mining algorithms– Information source is dynamic– System “pushes” information to user– Applications: news filter, recommender systems
3
2010 © University of Michigan
Examples of Information Filtering• News filtering• Email filtering• Movie/book recommenders• Literature recommenders • And many others …
4
2010 © University of Michigan
Content-based Filtering vs. Collaborative Filtering
• Basic filtering question: Will user U like item X?• Two different ways of answering it
– Look at what U likes – Look at who likes X
• Can be combined
characterize X content-based filtering
characterize U collaborative filtering
5
2010 © University of Michigan
Content-based filtering is also called
1. “Adaptive Information Filtering” in TREC2. “Selective Dissemination of Information (SDI)in Library & Information ScienceCollaborative filtering is also called “Recommender Systems”
6
2010 © University of Michigan
Why Bother?• Why do we need recommendation?• Why should amazon provide recommendation?• Why does Netflix spend millions to improve their
recommendation algorithm?
• Why can’t we rely on a search engine for all the recommendation?
7
2010 © University of MichiganChris Anderson, ‘The Long Tail’, Wired, Issue 12.10 - October 2004
Need a way to discover items that match our interests & tastes among tens or hundreds of thousands
The Long Tail
8
2010 © University of Michigan
Part I: Adaptive Filtering
9
2010 © University of Michigan
Adaptive Information Filtering• Stable & long term interest, dynamic info. source• System must make a delivery decision
immediately as a document “arrives”
FilteringSystem…
my interest:
10
2010 © University of Michigan
AIF vs. Retrieval & Categorization• Like retrieval over a dynamic stream of docs, but
(global) ranking is impossible• Like online binary categorization, but with no initial
training data and with limited feedback• Typically evaluated with a utility function
– Each delivered doc gets a utility value– Good doc gets a positive value– Bad doc gets a negative value– E.g., Utility = 3* #good - 2 *#bad (linear utility)
11
2010 © University of Michigan
A Typical AIF System
...BinaryClassifierUserInterestProfile
User
Doc Source
Accepted Docs
Initialization
Learning FeedbackAccumulated Docs
utility func.
User profile text
12
2010 © University of Michigan
Three Basic Problems in AIF• Making filtering decision (Binary classifier)
– Doc text, profile text yes/no• Initialization
– Initialize the filter based on only the profile text or very few examples
• Learning from– Limited relevance judgments (only on “yes” docs)– Accumulated documents
• All trying to maximize the utility
13
2010 © University of Michigan
Major Approaches to AIF• “Extended” retrieval systems
– “Reuse” retrieval techniques to score documents– Use a score threshold for filtering decision– Learn to improve scoring with traditional feedback– New approaches to threshold setting and learning
• “Modified” categorization systems– Adapt to binary, unbalanced categorization– New approaches to initialization– Train with “censored” training examples
14
2010 © University of Michigan
Difference between Ranking, Classification, and Recommendation
• What is the query?– Ranking: no query– Classification: labeled data (a.k.a, training data)– Recommendation: user + items
• What is the target?– Ranking: order of objects– Classification: labels of objects– Recommendation: Score? Label? Order?
15
2010 © University of Michigan
A General Vector-Space Approach
doc vector
profile vector
Scoring Thresholding
yes
no
FeedbackInformation
VectorLearning
ThresholdLearning
threshold
UtilityEvaluation
16
2010 © University of Michigan
Difficulties in Threshold Learning
36.5 R33.4 N32.1 R29.9 ?27.3 ?…...
=30.0
• Censored data• Little/none labeled data• Scoring bias due to
vector learning• Exploration vs.
Exploitation
17
2010 © University of Michigan
Threshold Setting in Extended Retrieval Systems
• Utility-independent approaches (generally not working well, not covered in this lecture)
• Indirect (linear) utility optimization– Logistic regression (score prob. of relevance)
• Direct utility optimization– Empirical utility optimization – Expected utility optimization given score distributions
• All try to learn the optimal threshold
18
2010 © University of Michigan
Logistic Regression (Robertson & Walker. 00)
• General idea: convert score of D to p(R|D)
• Fit the model using feedback data• Linear utility is optimized with a fixed prob. cutoff• But,
– Possibly incorrect parametric assumptions– No positive examples initially– Censored data and limited positive feedback– Doesn’t address the issue of exploration
)()|(log DsDRO
19
2010 © University of Michigan
Direct Utility Optimization• Given
– A utility function U(CR+ ,CR- ,CN+ ,CN-)– Training data D={<si, {R,N,?}>}
• Formulate utility as a function of the threshold and training data: U=F(,D)
• Choose the threshold by optimizing F(,D), i.e.,
),(maxarg DF
20
2010 © University of Michigan
Empirical Utility Optimization• Basic idea
– Compute the utility on the training data for each candidate threshold (score of a training doc)
– Choose the threshold that gives the maximum utility• Difficulty: Biased training sample!
– We can only get an upper bound for the true optimal threshold.
• Solutions:– Heuristic adjustment(lowering) of threshold – Lead to “beta-gamma threshold learning”
21
2010 © University of Michigan
Score Distribution Approaches( Aramptzis & Hameren 01; Zhang & Callan 01)
• Assume generative model of scores p(s|R), p(s|N)
• Estimate the model with training data • Find the threshold by optimizing the expected
utility under the estimated model• Specific methods differ in the way of defining
and estimating the scoring distributions
22
2010 © University of Michigan
Gaussian-Exponential Distributions• P(s|R) ~ N(,2) p(s-s0|N) ~ E()
)()(
)|()|( 02
2
021 ss
s
eNsspeRsp
(From Zhang & Callan 2001)23
2010 © University of Michigan
Score Distribution Approaches (cont.)
• Pros– Principled approach– Arbitrary utility– Empirically effective
• Cons– May be sensitive to the scoring function– Exploration not addressed
24
2010 © University of Michigan
Part II: Collaborative Filtering
25
2010 © University of Michigan
What is Collaborative Filtering (CF)?
• Making filtering decisions for an individual user based on the judgments of other users
• Inferring individual’s interest/preferences from that of other similar users
• General idea– Given a user u, find similar users {u1, …, um}– Predict u’s preferences based on the preferences of
u1, …, um
26
2010 © University of Michigan
CF: Assumptions• Users with a common interest will have similar
preferences • Users with similar preferences probably share the
same interest• Examples
– “interest is IR” “favor SIGIR papers”– “favor SIGIR papers” “interest is IR”
• Sufficiently large number of user preferences are available
27
2010 © University of Michigan
CF: Intuitions• User similarity (Paul Resnick vs. Rahul Sami)
– If Paul liked the paper, Rahul will like the paper– ? If Paul liked the movie, Rahul will like the movie– Suppose Paul and Rahul viewed similar movies in the
past six months …
• Item similarity– Since 90% of those who liked Star Wars also liked
Independence Day, and, you liked Star Wars– You may also like Independence Day
The content of items “didn’t matter”!28
2010 © University of Michigan
Rating-based vs. Preference-based• Rating-based: User’s preferences are
encoded using numerical ratings on items– Complete ordering– Absolute values can be meaningful – But, values must be normalized to combine
• Preference-based: User’s preferences are represented by partial ordering of items (Learning to Rank!)– Partial ordering– Easier to exploit implicit preferences
29
2010 © University of Michigan
A Formal Framework for Rating
u1
u2
…
ui
...
um
Users: U
Objects: O
o1 o2 … oj … on
3 1.5 …. … 2
2
1
3
Xij=f(ui,oj)=?
?
The task
Unknown function f: U x O R
• Assume known f values for some (u,o)’s
• Predict f values for other (u,o)’s
• Essentially function approximation, like other learning problems
30
2010 © University of Michigan
Where are the intuitions?• Similar users have similar preferences
– If u u’, then for all o’s, f(u,o) f(u’,o)• Similar objects have similar user preferences
– If o o’, then for all u’s, f(u,o) f(u,o’)• In general, f is “locally constant”
– If u u’ and o o’, then f(u,o) f(u’,o’)– “Local smoothness” makes it possible to predict
unknown values by interpolation or extrapolation• What does “local” mean?
31
2010 © University of Michigan
Two Groups of Approaches• Memory-based approaches
– f(u, o) = g(u)(o) g(u’)(o) if u u’ (g =preference function)
– Find “neighbors” of u and combine g(u’)(o)’s • Model-based approaches (not covered)
– Assume structures/model: object cluster, user cluster, f’ defined on clusters
– f(u, o) = f’(cu, co)– Estimation & Probabilistic inference
32
2010 © University of Michigan
Memory-based Approaches (Resnick et al. 94)
• General ideas:– xij: rating of object j by user i– ni: average rating of all objects by user i– Normalized ratings: – Memory-based prediction
• Specific approaches differ in w(a, i) -- the distance/similarity between user a and i
m
iijdj viawkv
1
),(
m
i
iawk1
),(/1 aajaj nvx
iijij nxv
33
2010 © University of Michigan
User Similarity Measures• Pearson correlation coefficient (sum over
commonly rated items)
• Cosine measure
• Many other possibilities!
jiij
jaaj
jiijaaj
pnxnx
nxnxiaw
22 )()(
))((),(
n
jij
n
jaj
n
jijaj
c
xx
xxiaw
1
2
1
2
1),(
34
2010 © University of Michigan
Improving User Similarity Measures
(Breese et al. 98)
• Dealing with missing values: default ratings• Inverse User Frequency (IUF): similar to IDF• Case Amplification: use w(a, i)p, e.g., p=2.5
35
2010 © University of Michigan
A Network Interpretation• If your neighbors love it,
you are likely to love it• Closer neighbors make
a larger impact• Measure the closeness
by the items you and her liked/disliked
36
2010 © University of Michigan
Network based approaches• No clear notion of “distance”
now, but the “scores” are estimated based on some propagation process.
• Score propagation based on random walk
• Two approaches:– Starting from the user, how
likely/how long can I reach the object?
– Starting from the object, how likely/how long can I hit the user?
37
2010 © University of Michigan
Random Walk and Hitting Time
i
kA
jP = 0.7
P = 0.3
• Hitting Time– TA: the first time that the random
walk is at a vertex in A• Mean Hitting Time
– hiA: expectation of TA given that
the walk starts from vertex i
0.3
0.7
38
2010 © University of Michigan
Computing Hitting Time
i
kA
j
TA: the first time that the random walk is at a vertex in A
}0,:min{ tAXtT tA
A ifor ,1)( Vj
Ajhjip
Aih
A ifor ,0
Iterative Computation
hiA: expectation of TA given that the
walk starting from vertex i
A i
h = 0
hiA = 0.7 hj
A + 0.3 hkA + 1
0.7
0.7
Apparently, hiA = 0 for those
39
2010 © University of Michigan
Bipartite Graph and Hitting Time
Expected proximity of query i to the query A : hitting time of i A, hi
A
Bipartite Graph:- Edges between V1 and V2
- No edge inside V1 or V2
- Edges are weighted- e.g., V1 = query; V2 = Url
A
ijw(i, j) = 3
4
5
0.7
0.4V1 V2
7 1
)73(3),()(
idjiwjip
A
ij
4
5
0.7
0.4V1 V2
7 1
)13(3),()(
jdjiwijp
A
k
ij
4
5
0.7
0.4V1 V2
7 1
2
),(),()(Vj ji d
jkwd
jiwkip
• convert to a directed graph, even collapse one group
40
2010 © University of Michigan
Example: Query Suggestion
Welcome to the hotel california
Suggestions
hotel california
eagles hotel california
hotel california band
hotel california by the eagles
hotel california song
lyrics of hotel california
listen hotel california eagle
2010 © University of Michigan
Generating Query Suggestion using Click Graph (Mei et al. 08)
42
Aaa
american airline
mexiana
www.aa.com
www.theaa.com/travelwatch/planner_main.jsp
en.wikipedia.org/wiki/Mexicana
300
15
Query Url• Construct a (kNN)
subgraph from the query log data (of a predefined number of queries/urls)
• Compute transition probabilities p(i j)
• Compute hitting time hiA
• Rank candidate queries using hi
A
freq(q, url)
2010 © University of Michigan
Result: Query Suggestion
Hitting time
wikipedia friends
friends tv show wikipedia
friends home page
friends warner bros
the friends series
friends official site
friends(1994)
friendship
friends poem
friendster
friends episode guide
friends scripts
how to make friends
true friends
Yahoo
secret friends
friends reunited
hide friends
hi 5 friends
find friends
poems for friends
friends quotes
Query = friends
43
2010 © University of Michigan
Summary• Filtering is “easy”
– The user’s expectation is low – Any recommendation is better than none– Making it practically useful
• Filtering is “hard”– Must make a binary decision, though ranking is
also possible– Data sparseness– “Cold start”
44
2010 © University of Michigan
Example: NetFlix
45
2010 © University of Michigan46
References (Adaptive Filtering)· General papers on TREC filtering evaluation
· D. Hull, The TREC-7 Filtering Track: Description and Analysis , TREC-7 Proceedings. · D. Hull and S. Robertson, The TREC-8 Filtering Track Final Report, TREC-8 Proceedings. · S. Robertson and D. Hull, The TREC-9 Filtering Track Final Report, TREC-9 Proceedings.· S. Robertson and I. Soboroff, The TREC 2001 Filtering Track Final Report, TREC-10
Proceedings
· Papers on specific adaptive filtering methods · Stephen Robertson and Stephen Walker (2000), Threshold Setting in Adaptive Filtering .
Journal of Documentation, 56:312-331, 2000
· Chengxiang Zhai, Peter Jansen, and David A. Evans, Exploration of a heuristic approach to threshold learning in adaptive filtering, 2000 ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR'00), 2000. Poster presentation.
· Avi Arampatzis and Andre van Hameren The Score-Distributional Threshold Optimization for Adaptive Binary Classification Tasks , SIGIR'2001.
· Yi Zhang and Jamie Callan, 2001, Maximum Likelihood Estimation for Filtering Threshold, SIGIR 2001.
· T. Ault and Y. Yang, 2002, kNN, Rocchio and Metrics for Information Filtering at TREC-10, TREC-10 Proceedings.
2010 © University of Michigan
References (Collaborative Filtering)
• Resnick, P., Iacovou, N., Suchak, M., Bergstrom, P., and Riedl, J., "GroupLens: An Open Architecture for Collaborative Filtering of Netnews," CSCW. 175-186.
• P Resnick, HR Varian, Recommender Systems, CACM 1997
• Cohen, W.W., Schapire, R.E., and Singer, Y. (1999) "Learning to Order Things", Journal of AI Research, Volume 10, pages 243-270.
• Freund, Y., Iyer, R.,Schapire, R.E., & Singer, Y. (1999). An efficient boosting algorithm for combining preferences. Machine Learning Journal. 1999.
• Breese, J. S., Heckerman, D., and Kadie, C. (1998). Empirical Analysis of Predictive Algorithms for Collaborative Filtering. In Proceedings of the 14th Conference on Uncertainty in Articial Intelligence, pp. 43-52.
• Alexandrin Popescul and Lyle H. Ungar, Probabilistic Models for Unified Collaborative and Content-Based Recommendation in Sparse-Data Environments, UAI 2001.
• N. Good, J.B. Schafer, J. Konstan, A. Borchers, B. Sarwar, J. Herlocker, and J. Riedl. "Combining Collaborative Filtering with Personal Agents for Better Recommendations." Proceedings AAAI-99. pp 439-446. 1999.
• Qiaozhu Mei, Dengyong Zhou, Kenneth Church. Query Suggestion Using Hitting Time, CIKM'08, pages 469-478
47