Information Extraction

Information Extraction

Yunyao Li

EECS /SI 76703/29/2006

The ProblemDate

Time: Start - End

Speaker

Person

Location

What is “Information Extraction”Filling slots in a database from sub-segments of text.As a task:

October 14, 2002, 4:00 a.m. PT

For years, Microsoft Corporation CEO Bill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation.

Today, Microsoft claims to "love" the open-source concept, by which software code is made public to encourage improvement and development by outside programmers. Gates himself says Microsoft will gladly disclose its crown jewels--the coveted code behind the Windows operating system--to select customers.

"We can be open source. We love the concept of shared source," said Bill Veghte, a Microsoft VP. "That's a super-important shift for us in terms of code access.“

Richard Stallman, founder of the Free Software Foundation, countered saying…

NAME TITLE ORGANIZATION

Courtesy of William W. Cohen

What is “Information Extraction”Filling slots in a database from sub-segments of text.As a task:


October 14, 2002, 4:00 a.m. PT





NAME TITLE ORGANIZATIONBill Gates CEO MicrosoftBill Veghte VP MicrosoftRichard Stallman founder Free Soft..

IE

What is “Information Extraction”


Information Extraction = segmentation + classification + association + clustering

October 14, 2002, 4:00 a.m. PT





Microsoft CorporationCEOBill GatesMicrosoftGatesMicrosoftBill VeghteMicrosoftVPRichard StallmanfounderFree Software Foundation

aka “named entity extraction”




October 14, 2002, 4:00 a.m. PT








October 14, 2002, 4:00 a.m. PT









October 14, 2002, 4:00 a.m. PT






Microsoft CorporationCEOBill GatesMicrosoftGatesMicrosoftBill VeghteMicrosoftVPRichard StallmanfounderFree Software Foundation NA

ME

TITL

E

ORGA

NIZA

TION

Bill

Gat

esCEO

Micr

osof

tBill

Ve g

h te

VPMicr

osof

tRich

ard

Stal

lman

foun

der

Free

Sof

t..

*

*

*

*

Live Example: Seminar

http://l2r.cs.uiuc.edu/~cogcomp/sem_demo.php

Landscape of IE Techniques


Lexicons

AlabamaAlaska…WisconsinWyoming

Abraham Lincoln was born in Kentucky.

member?

Classify Pre-segmentedCandidates

Abraham Lincoln was born in Kentucky.

Classifier

which class?

Sliding WindowAbraham Lincoln was born in Kentucky.

Classifierwhich class?

Try alternatewindow sizes:

Boundary ModelsAbraham Lincoln was born in Kentucky.

Classifier

which class?

BEGIN END BEGIN END

BEGIN

Context Free GrammarsAbraham Lincoln was born in Kentucky.

NNP V P NPVNNP

NP

PP

VPVP

S

Mos

t lik

ely

pars

e?

Finite State MachinesAbraham Lincoln was born in Kentucky.

Most likely state sequence?

Our Focus today!

Markov Property

S2

S2S1

1/2

1/2 1/3

2/3

1

The state of a system at time t+1, qt+1, is conditionally independent of {qt-1, qt-2, …, q1, q0} given qt

In another word, current state determines the probability distribution for the next state.

S1: rainS2: cloudS3: sun

Markov Property

S2

S3S1

1/2

1/2 1/3

2/3

1

State-transition probabilities,

A =

S1: rainS2: cloudS3: sun

033.067.005.05.0100

Q: given today is sunny (i.e., q1=3),what is the probability of “sun-cloud”with the model?

Hidden Markov ModelS1: rainS2: cloudS3: sun

S2

S3S1

1/2

1/2 1/3

2/3

14/5

1/10

7/101/5 3/10

9/10

observations

O1 O2 O3 O4 O5

state sequences

IE with Hidden Markov ModelSI/EECS 767 is held weekly at SIN2 .

SI/EECS 767 is held weekly at SIN2

Course name: SI/EECS 767

Given a sequence of observations:

and a trained HMM:

Find the most likely state sequence: (Viterbi)

Any words said to be generated by the designated “course name”state extract as a course name:

),(maxarg osPs

course namelocation namebackground

Name Entity Extraction[Bikel, et al 1998]

Person

Org

Other

(Five other name classes)

start-of-sentence

end-of-sentence

Hidden states

Name Entity ExtractionTransitionprobabilities

Observationprobabilities

P(st | st-1, ot-1 ) P(ot | st , st-1 )

P(ot | st , ot-1 )or

(1) Generating first word of a name-class

(2) Generating the rest of words in the name-class

(3) Generating “+end+” in a name-class

Training: Estimating Probabilities

Back-Off“unknown words” and insufficient training data

P(st | st-1 )P(st )

P(ot | st )P(ot )

Transitionprobabilities

Observationprobabilities

HMM-Experimental Results

Train on ~500k words of news wire text.

Results:

Learning HMM for IE[Seymore, 1999]

Consider labeled, unlabeled, and distantly-labeled data

Some Issues with HMM• Need to enumerate all possible observation

sequences• Not practical to represent multiple interacting

features or long-range dependencies of the observations

• Very strict independence assumptions on the observations

Maximum Entropy Markov Models

S t -1 S t

O t

S t+1

O t +1Ot -1

identity of wordends in “-ski”is capitalizedis part of a noun phraseis in a list of city namesis under node X in WordNetis in bold fontis indentedis in hyperlink anchor…

…

…part of

noun phrase

is “Wisniewski”

ends in “-ski”

Idea: replace generative model in HMM with a maxent model, where state depends on observations

...)|Pr( tt xsCourtesy of William W. Cohen

[Lafferty, 2001]

MEMMS t -1 S t

O t

S t+1

O t +1Ot -1

identity of wordends in “-ski”is capitalizedis part of a noun phraseis in a list of city namesis under node X in WordNetis in bold fontis indentedis in hyperlink anchor…

…

…part of

noun phrase

is “Wisniewski”

ends in “-ski”

Idea: replace generative model in HMM with a maxent model, where state depends on observations and previous state history

......),|Pr( ,2,1 tttt ssxsCourtesy of William W. Cohen

HMM vs. MEMMSt-1 St

Ot

St+1

Ot+1Ot-1

...

iiiii sossos )|Pr()|Pr(),Pr( 11

St-1 St

Ot

St+1

Ot+1Ot-1

...

iiii ossos ),|Pr()|Pr( 11

Label Bias Problem with MEMM

Consider this MEMM

Pr(12|ro) = Pr(2|1,ro)Pr(1,ro) = Pr(2| 1,o)Pr(1,r)

Pr(2|1,o) = Pr(2|1,r) = 1

Pr(12|ri) = Pr(2|1,ri)Pr(1,ri) = Pr(2| 1,i)Pr(1,r)

Pr(12|ro) = Pr(12|ri)

But it should be Pr(12|ro) < Pr(12|ri)!

Solve the Label Bias Problem• Change the state-transition structure of the model

– Not always practical to change the set of states

• Start with a fully-connected model and let the training procedure figure out a good structure– Prelude the use of prior, which is very valuable (e.g.

in information extraction)

Random Field

Courtesy of Rongkun Shen

Conditional Random Field

Courtesy of Rongkun Shen

Conditional Distribution

1 2 1 2( , , , ; , , , ); andn n k k

x is a data sequencey is a label sequence v is a vertex from vertex set V = set of label random variablese is an edge from edge set E over Vfk and gk are given and fixed. gk is a Boolean vertex feature; fk

is a Boolean edge featurek is the number of features

are parameters to be estimated

y|e is the set of components of y defined by edge ey|v is the set of components of y defined by vertex v

If the graph G = (V, E) of Y is a tree, the conditional distribution over the label sequence Y = y, given X = x, by fundamental theorem of random fields is:

(y | x) exp ( , y | , x) ( , y | , x)

k k e k k v

e E,k v V ,k

p f e g v

Conditional Distribution• CRFs use the observation-dependent normalization Z(x) for the conditional distributions:

Z(x) is a normalization over the data sequence x

(y | x) exp ( , y | , x) ( , y |1(x)

, x)

k k e k k v

e E,k v V ,k

p f e g vZ

HMM like CRF Single feature for each state-state pair (y’,y) and state-

observation pair in the data to train CRF

Yt-1 Yt

Xt

Yt+1

Xt+1Xt-1

...=

01 if yu = y’ and yv = y

otherwise

= 01 if yv = y and xv = x

otherwise

y’,y and µy,x are equivalent to the logarithm of the HMM transition probability Pr(y’|y) and observation probability Pr(x|y)

HMM like CRF For a chain structure, the conditional probability of a label

sequence can be expressed in matrix form. For each position i in the observed sequence x, define matrix

Where ei is the edge with label (yi-1, yi) and vi is the vertex with label yi

HMM like CRFThe normalization function is the (start, stop) entry of the productof these matrices

The conditional probability of label sequence y is:

where, y0 = start and yn+1 = stop

Parameter EstimationThe problem: determine the parameters From training data with empirical distribution

The goal: maximize the log-likelihood objective function

Parameter Estimation – Iterative Scaling Algorithms

Update the weights as and for Appropriately chosen

for edge feature fk is the solution of

T(x, y) is a global property of (x,y) and efficiently computing the Right-hand sides of the above equation is a problem

Algorithm SDefine slack feature:

For each index i = 0, …, n+1 we define forward vectors

And backward vectors

p y y( ' , )

Algorithm S=

= p x p y y x f e y xx n

n

k i e y yy y

i

( ) ( ' , , ) ( , | , )( ', )'1

p y y xy x M y y x y x

Z xi i i( ' , , )

( ' | ) ( ' , | ) ( | )( )

1

=

Algorithm S

The rate of convergence is governed by step size which is Inversely proportional to constant S, but S is generally quite large, resulting in slow convergence.

Algorithm TKeeps track of partial T total. It accumulates feature expectations into counters indexed by T(x)

Use forward-back ward recurrences to compute the expectationak,t of feature fk and bk,t of feature gk given that T(x) = t

Experiments• Modeling label bias problem

– 2000 training and 500 test samples generated by HMM

– CRF error is 4.6%– MEMM error is 42%

CRF solves label bias problem

Experiments• Modeling mixed order sources

– CRF converge in 500 iterations– MEMM converge in 100 iterations

MEMM vs. HMMThe HMM outperforms the MEMM

CRF vs. MEMMCRF usually outperforms the MEMM

CRF vs. HMMEach open square represents a data set with < ½, and a sold square indicates a data set with a ½. When the data is mostly second order ½, the discriminatively trained CRF usually outperforms the MEMM

POS Tagging Experiments• First-order HMM, MEMM and CRF model• Data set: Penn Tree bank• 50-50% test-train split

• Uses MEMM parameter vector as a starting point for training the corresponding CRF to accelerate convergence speed.

Interactive IE using CRF

Interactive parser updates IE results according to user’s changes. Color coding used to alert the ambiguity of IE.

Some IE tools Available• MALLET (UMass)

– statistical natural language processing, – document classification, – clustering, – information extraction– other machine learning applications to text.

• Sample Application: GeneTaggerCRF: a gene-entity tagger based on

MALLET (MAchine Learning for LanguagE Toolkit). It uses conditional random fields to find genes in a text file.

• http://minorthird.sourceforge.net/• “a collection of Java classes for storing

text, annotating text, and learning to extract entities and categorize text”

• Stored documents can be annotated in independent files using TextLabels (denoting, say, part-of-speech and semantic information)

MinorThird

http://minorthird.sourceforge.net/

GATE• http://gate.ac.uk/ie/annie.html• leading toolkit for Text Mining • distributed with an Information Extraction

component set called ANNIE (demo)• Used in many research projects

– Long list can be found on its website– Under integration of IBM UIMA

http://gate.ac.uk/ie/annie.html

http://gate.ac.uk/ie/index.html

http://gate.ac.uk/annie/

Sunita Sarawagi's CRF package

• http://crf.sourceforge.net/• A Java implementation of conditional

random fields for sequential labeling.

http://crf.sourceforge.net/

UIMA (IBM) • Unstructured Information

Management Architecture.– A platform for unstructured

information management solutions from combinations of semantic analysis (IE) and search components.

Some Interesting Website based on IE

• ZoomInfo• CiteSeer.org (some of us using it everyday!)

• Google Local, Google Scholar• and many more…

http://www.zoominfo.com/

http://www.citeseer.org/

http://local.google.com/

http://scholar.google.com/

Information Extraction

Documents