Download Massimo Poesio: Text Categorization and

CEEC 2015 TUTORIAL: TEXT ANALYTICS Uni Essex Kruschwitz, Poesio, AlthobaiG Language & ComputaGon Group THE PLAN FOR THE TUTORIAL •  14-­‐15: Lecture 1, Intro to NLP & InformaGon Retrieval (Kruschwitz) •  15-­‐16: Lecture 2, Text Mining (Poesio) •  16-­‐16:30: Coffee break •  16:30-­‐17:30: Lab, SenGment Analysis (Poesio) •  17:30-­‐18:30: Lab, GATE/NER (AlthobaiG) WEB PAGE •  h`p://csee.essex.ac.uk/staff/poesio/Teach/
TextAnalyGcsTutorial/ Text AnalyGcs Massimo Poesio Lecture 2: Machine Learning in NLP / ClassificaGon / SenGment Analysis / Stylometry / InformaGon ExtracGon APPLICATIONS OF TEXT ANALYTICS •  Text analyGcs techniques are widely used these days, for at least two reasons: 1.  The explosion of the Web in general and of social media in parGcular, and the increasing shie to digital documents, have enormously increased both the need to manage these textual data and the desire to take advantage of the opportunity 2.  For a number of these tasks, decent results can be obtained using methods that do not require high-­‐performance linguisGc processing EXAMPLE: IS THIS SPAM? From: "" <[email protected]> Subject: real estate is the only way... gem oalvgkay Anyone can buy real estate with no money down Stop paying rent TODAY ! There is no need to spend hundreds or even thousands for similar courses I am 22 years old and I have already purchased 6 properGes using the methods outlined in this truly INCREDIBLE ebook. Change your life NOW ! ===========================================
====== Click Below to order: h`p://www.wholesaledaily.com/sales/nmd.htm ===========================================
====== Dear Hamming Seminar Members
The next Hamming Seminar will take place on
Wednesday 25th May and the details are as follows Who: Dave Robertson Title: Formal Reasoning Gets
Social Abstract: For much of its history, formal knowledge
representation has aimed to describe knowledge
independently of the personal and social context in which
it is used, with the advantage that we can automate
reasoning with such knowledge using mechanisms that
also are context independent. This sounds good until you
try it on a large scale and find out how sensitive to
context much of reasoning actually is. Humans, however,
are great hoarders of information and sophisticated tools
now make the acquisition of many forms of local
knowledge easy. The question is: how to combine this
beyond narrow individual use, given that knowledge (and
reasoning) will inevitably be contextualised in ways that
may be hidden from the people/systems that may
interact to use it? This is the social side of knowledge
representation and automated reasoning. I will discuss
how the formal reasoning community has adapted to this
new view of scale. When: 4pm, Wednesday 25 May 2011
Where: Room G07, Informatics Forum There will be wine
and nibbles afterwards in the atrium café area.
EXAMPLE: IS THIS SPAM? Palm Garden Hotel E-Newsletter February, 2013
Aroi Dee Thai Restaurant Tel: (603) 8943 2233
Tantalising Thai Cuisine
Love Thai Cuisine? A visit to Aroi Dee Thai Restaurant is a must. With our team of
talented and experienced Thai Chefs, taste the tantalising and authentic dishes they
prepare. A selection of Nyonya cuisine also available ... more
Aroi Dee Thai Restaurant
Tel: (603) 8943 2233
Chinese New Year Set Menus on 28 Jan - 24 Feb
A 9 course menu featuring Prosperity Combination Yee Sang and Jelly Fish as the
starter. This is followed by Braised Assorted Seafood Soup with Beancurd and Fish
Roe, Roasted Crispy Chicken Cantonese Style and Steamed Red Mullet Fish with
Light Soya Sauce. Continue the meal with Wok Fried Prawns with Marmite Sauce,
Braised Mushroom with Beancurd and Broccoli and Steamed Lotus Leaf Rice with
Yam and Assorted Preserved Meat. Complete the meal with refreshing desserts of
Chilled Sea Coconut and Sno more
THE ROLE OF MACHINE LEARNING •  In modern text analyGcs, instead of giving the program an algorithm to do a task, we give an algorithm for learning how to do the task •  Specifically, given a set of examples, the system learns a funcGon that does a good job of expressing the relaGonship: –  Categorizing email messages as a funcGon from emails to their category (spam, useful) –  A checker playing strategy a funcGon from moves to their values (winning, losing) SUPERVISED AND UNSUPERVISED METHODS •  Both supervised and unsupervised ML is used in text analyGcs –  Supervised: spam detecGon, senGment analysis, informaGon extracGon –  Unsupervised: document clustering, summarizaGon CLASSIFICATION SPAM NON-­‐SPAM EXAMPLE: DECISION TREES •  A DECISION TREE is a classifier in the form of a tree structure, where each node is either a: –  Leaf node - indicates the value of the target
attribute (class) of examples, or
–  Decision node - specifies some test to be carried
out on a single attribute-value, with one branch
and sub-tree for each possible outcome of the
test.
•  A decision tree can be used to classify an example by starGng at the root of the tree and moving through it unGl a leaf node, which provides the classificaGon of the instance. Example of a Decision Tree Tid Refund Marital
Status
Taxable
Income Cheat
1
Yes
Single
125K
No
2
No
Married
100K
No
3
No
Single
70K
No
4
Yes
Married
120K
No
5
No
Divorced 95K
Yes
6
No
Married
No
7
Yes
Divorced 220K
No
8
No
Single
85K
Yes
9
No
Married
75K
No
10
No
Single
90K
Yes
60K
Splitting Attributes
Refund
Yes
No
NO
MarSt
Single, Divorced
TaxInc
< 80K
NO
NO
> 80K
YES
10
Training Data
Married
Model: Decision Tree
Another Example of Decision Tree 10
Tid Refund Marital
Status
Taxable
Income Cheat
1
Yes
Single
125K
No
2
No
Married
100K
No
3
No
Single
70K
No
4
Yes
Married
120K
No
5
No
Divorced 95K
Yes
6
No
Married
No
7
Yes
Divorced 220K
No
8
No
Single
85K
Yes
9
No
Married
75K
No
10
No
Single
90K
Yes
60K
Married
MarSt
NO
Single,
Divorced
Refund
No
Yes
NO
TaxInc
< 80K
NO
> 80K
YES
There could be more than one tree that
fits the same data!
Decision Tree ClassificaGon Task Tid
Attrib1
Attrib2
Attrib3
Class
1
Yes
Large
125K
No
2
No
Medium
100K
No
3
No
Small
70K
No
4
Yes
Medium
120K
No
5
No
Large
95K
Yes
6
No
Medium
60K
No
7
Yes
Large
220K
No
8
No
Small
85K
Yes
9
No
Medium
75K
No
10
No
Small
90K
Yes
Tree
Induction
algorithm
Induction
Learn
Model
Model
10
Training Set
Tid
Attrib1
Attrib2
11
No
Small
55K
?
12
Yes
Medium
80K
?
13
Yes
Large
110K
?
14
No
Small
95K
?
15
No
Large
67K
?
10
Test Set
Attrib3
Apply
Model
Class
Deduction
Decision
Tree
Apply Model to Test Data Test Data
Start from the root of tree.
Refund
Yes
10
No
NO
MarSt
Single, Divorced
TaxInc
< 80K
NO
Married
NO
> 80K
YES
Refund Marital
Status
Taxable
Income Cheat
No
80K
Married
?
Apply Model to Test Data Test Data
Refund
Yes
10
No
NO
MarSt
Single, Divorced
TaxInc
< 80K
NO
Married
NO
> 80K
YES
Refund Marital
Status
Taxable
Income Cheat
No
80K
Married
?
Apply Model to Test Data Test Data
Refund
10
Yes
No
NO
MarSt
Single, Divorced
TaxInc
< 80K
NO
Married
NO
> 80K
YES
Refund Marital
Status
Taxable
Income Cheat
No
80K
Married
?
Apply Model to Test Data Test Data
Refund
10
Yes
No
NO
MarSt
Single, Divorced
TaxInc
< 80K
NO
Married
NO
> 80K
YES
Refund Marital
Status
Taxable
Income Cheat
No
80K
Married
?
Apply Model to Test Data Test Data
Refund
10
Yes
No
NO
MarSt
Single, Divorced
TaxInc
< 80K
NO
Married
NO
> 80K
YES
Refund Marital
Status
Taxable
Income Cheat
No
80K
Married
?
Apply Model to Test Data Test Data
Refund
No
NO
MarSt
Single, Divorced
TaxInc
NO
Taxable
Income Cheat
No
80K
Married
?
10
Yes
< 80K
Refund Marital
Status
Married
NO
> 80K
YES
Assign Cheat to “No”
Decision Tree ClassificaGon Task Tid
Attrib1
Attrib2
Attrib3
Class
1
Yes
Large
125K
No
2
No
Medium
100K
No
3
No
Small
70K
No
4
Yes
Medium
120K
No
5
No
Large
95K
Yes
6
No
Medium
60K
No
7
Yes
Large
220K
No
8
No
Small
85K
Yes
9
No
Medium
75K
No
10
No
Small
90K
Yes
Tree
Induction
algorithm
Induction
Learn
Model
Model
10
Training Set
Tid
Attrib1
Attrib2
11
No
Small
55K
?
12
Yes
Medium
80K
?
13
Yes
Large
110K
?
14
No
Small
95K
?
15
No
Large
67K
?
10
Test Set
Attrib3
Apply
Model
Class
Deduction
Decision
Tree
Decision Tree InducGon •  Many Algorithms: –  Hunt’s Algorithm (one of the earliest) –  CART –  ID3, C4.5 –  SLIQ,SPRINT WORD-­‐BASED METHODS •  In many cases, machine learning methods applied to text analyGcs tasks can achieve decent results relying only on the occurrence of WORDS –  Or on meta-­‐features easily extractable from a document IS THIS SPAM? From: "" <[email protected]> Subject: real estate is the only way... gem oalvgkay Anyone can buy real estate with no money down Stop paying rent TODAY ! There is no need to spend hundreds or even thousands for similar courses I am 22 years old and I have already purchased 6 properGes using the methods outlined in this truly INCREDIBLE ebook. Change your life NOW ! ================================================= Click Below to order: h`p://www.wholesaledaily.com/sales/nmd.htm ================================================= TEXT CATEGORIZATION •  Given: –  A descripGon of an instance, x∈X, where X is the instance language or instance space. •  Issue: how to represent text documents. –  A fixed set of categories: C = {c1, c2,…, cn} •  Determine: –  The category of x: c(x)∈C, where c(x) is a categoriza<on func<on whose domain is X and whose range is C. •  We want to know how to build categorizaGon funcGons (“classifiers”). Document ClassificaGon “planning!
language!
proof!
intelligence”!
Testing!
Data:!
(AI)!
(Programming)!
(HCI)!
Classes:!
ML!
Training!
Data:!
Planning!
Semantics!
Garb.Coll.!
Multimedia!
learning!
planning! programming! garbage!
...!
intelligence! temporal! semantics!
collection!
algorithm!
reasoning! language!
memory!
reinforcement! plan!
proof...!
optimization!
network...!
language...!
region...!
GUI!
...!
(Note: in real life there is oeen a hierarchy, not present in the above problem statement; and you get papers on ML approaches to Garb. Coll.) TEXT CLASSIFICATION WITH DT •  Build a separate decision tree for each category •  Use WORDS COUNTS as features Reuters Data Set (21578 -­‐ ModApte split) •  9603 training, 3299 test arGcles; ave. 200 words •  118 categories –  An arGcle can be in more than one category –  Learn 118 binary category disGncGons Common categories (#train, #test) •  Earn (2877, 1087)
•  Acquisitions (1650, 179)
•  Money-fx (538, 179)
•  Grain (433, 149)
•  Crude (389, 189)
•  Trade (369,119)
•  Interest (347, 131)
•  Ship (197, 89)
•  Wheat (212, 71)
•  Corn (182, 56)
28 AN EXAMPLE OF REUTERS TEXT FoundaGons of StaGsGcal Natural Language Processing, Manning and Schuetze Decision Tree for Reuter classificaGon FoundaGons of StaGsGcal Natural Language Processing, Manning and Schuetze SPAM CLASSIFICATION WITH DECISION TREES •  One of the best known spam detectors, SpamAssassin, was based on decision trees SpamAssassin Features 100
From: address is in the user's black-list
4.0
Sender is on www.habeas.com Habeas Infringer List
3.994
Invalid Date: header (timezone does not exist)
3.970
Written in an undesired language
3.910
Listed in Razor2, see http://razor.sf.net/
3.801
Subject is full of 8-bit characters
3.472
Claims compliance with Senate Bill 1618
3.437
exists:X-Precedence-Ref
3.371
Reverses Aging
3.350
Claims you can be removed from the list
3.284
'Hidden' assets
3.283
Claims to honor removal requests
3.261
Contains "Stop Snoring"
3.251
Received: contains a name with a faked IP-address
3.250
Received via a relay in list.dsbl.org
3.200
set indicates a foreign language
600.465
- Intro toCharacter
NLP - J.
Eisner
32 SpamAssassin Features 3.198
Forged eudoramail.com 'Received:' header found
3.193
Free Investment
3.180
Received via SBLed relay, seehttp://www.spamhaus.org/sbl/
3.140
Character set doesn't exist
3.123
Dig up Dirt on Friends
3.090
No MX records for the From: domain
3.072
X-Mailer contains malformed Outlook Expressversion
3.044
Stock Disclaimer Statement
3.009 Apparently, NOT Multi Level Marketing
3.005
Bulk email software fingerprint (jpfree) found inheaders
2.991
exists:Complain-To
2.975
Bulk email software fingerprint (VC_IPA) found inheaders
2.968
Invalid Date: year begins with zero
2.932
Mentions Spam law "H.R. 3113"
2.900
Received forged, contains fake AOL relays
2.879
600.465
- Intro toAsks
NLP - J. for credit card details
Eisner
33 SpamAssassin Features 2.858
To: username at front of subject
2.851
Claims you actually asked for this spam
2.842
To header contains 'recipient' marker
2.826
Compare Rates
2.800
Received: says mail bounced all around the world
2.800
Mentions Spam Law "UCE-Mail Act"
2.796
Received via buggy SMTP server (MDaemon2.7.4SP4R)
2.795
Bulk email software fingerprint (StormPost) foundin headers
2.786
Broken CGI script message
2.784
Message-Id generated by a spam tool
2.783
Urges you to call now
2.782
Tells you it's an ad
2.782
RAND found, spammer forgot to run the random-IDgenerator
2.748
Cable Converter
2.744
No Age Restrictions
2.737
porn - Celebrity Porn
600.465
- Intro toPossible
NLP - J.
Eisner
34 SpamAssassin Features 2.782
Tells you it's an ad
2.782
RAND found, spammer forgot to run the random-IDgenerator
2.748
Cable Converter
2.744
No Age Restrictions
2.737
Possible porn - Celebrity Porn
2.735
Bulk email software fingerprint (JiXing) found inheaders
2.730
DNSBL: sender is Confirmed Spam Source
2.726
Bulk email software fingerprint (MMailer) found inheaders
2.720
exists:X-Encoding
2.720
DNSBL: sender is Confirmed Open Relay
2.702
SEC-mandated penny-stock warning -- thanks SEC
2.695
Claims you can be removed from the list
2.693
Removes Wrinkles
2.668
Offers a stock alert
2.660
Listed in DCC, seehttp://rhyolite.com/anti-spam/dcc/
2.658
pyramid scheme phrase (1)
600.465
- Intro toCommon
NLP - J.
Eisner
35 OTHER LEARNING METHODS USED FOR TEXT CATEGORIZATION •  Bayesian methods (Naïve Bayes) •  Neural nets (e.g. ,perceptron) •  Vector-­‐space methods (k-­‐NN, Rocchio, unsupervised) •  SVMs PRACTICAL CORNER: ML TOOLS •  These days one doesn’t need to implement one’s own machine learning algorithms, many freely available, open source playorms exist •  Best known: WEKA –  Supports most best known ML algorithms –  Easy to use graphical interface –  Can be downloaded from h`p://www.cs.waikato.ac.nz/ml/weka/ 22/09/2015 University of Waikato 38 INPUT FILES @relaGon heart-­‐disease-­‐simplified @a`ribute age numeric @a`ribute sex { female, male} @a`ribute chest_pain_type { typ_angina, asympt, non_anginal, atyp_angina} @a`ribute cholesterol numeric @a`ribute exercise_induced_angina { no, yes} @a`ribute class { present, not_present} @data 63,male,typ_angina,233,no,not_present 67,male,asympt,286,yes,present 67,male,asympt,229,yes,present 38,female,non_anginal,?,no,not_present ... 22/09/2015 39 Supervised vs. unsupervised learning •  The setup for document classificaGon we just saw is called supervised learning in Machine Learning •  In the domain of text, various names –  Text classificaGon, text categorizaGon –  Document classificaGon/categorizaGon –  “AutomaGc” categorizaGon –  RouGng, filtering … •  In some cases however we don’t know the classes in advance •  In this case we talk of unsupervised learning –  Presumes no availability of training samples –  Clusters output may not be themaGcally unified. Unsupervised methods for text analyGcs: (document) clustering •  Clustering = discovering similariGes between objects –  Individuals, Documents, … •  ApplicaGons: –  Recommender systems –  Document organizaGon Recommending: restaurants •  We have a list of all Wivenhoe restaurants –  with ↑ and ↓ raGngs for some –  as provided by some Uni Essex students / staff •  Which restaurant(s) should I recommend to you? Input Algorithm 0 •  Recommend to you the most popular restaurants –  say # posiGve votes minus # negaGve votes •  Ignores your culinary preferences –  And judgements of those with similar preferences •  How can we exploit the wisdom of “like-­‐
minded” people? Another look at the input -­‐ a matrix Now that we have a matrix View all other entries as zeros for now.
PREFERENCE-­‐DEFINED DATA SPACE . . . . . . . . . . . . . . . . 47
Similarity between two people •  Similarity between their preference vectors. •  Inner products are a good start. •  Dave has similarity 3 with EsGe –  but -­‐2 with Cindy. •  Perhaps recommend Black Buoy to Dave –  and Bakehouse to Bob, etc. Algorithm 1.1 •  You give me your preferences and I need to give you a recommendaGon. •  I find the person “most similar” to you in my database and recommend something he likes. •  Aspects to consider: –  No a`empt to discern cuisines, etc. –  What if you’ve been to all the restaurants he has? –  Do you want to rely on one person’s opinions? Algorithm 1.k •  You give me your preferences and I need to give you a recommendaGon. •  I find the k people “most similar” to you in my database and recommend what’s most popular amongst them. •  Issues: –  A priori unclear what k should be –  Risks being influenced by “unlike minds” Slightly more sophisGcated a`empt •  Group similar users together into clusters •  You give your preferences and seek a recommendaGon, then –  Find the “nearest cluster” (what’s this?) –  Recommend the restaurants most popular in this cluster •  Features: –  avoids data sparsity issues –  sGll no a`empt to discern why you’re recommended what you’re recommended –  how do you cluster? CLUSTERS •  Can cluster Cindy Alice, Bob Fred .. Dave, EsGe …. DOCUMENT CLUSTERING •  Consider clustering a large set of computer science documents –  what do you expect to see in the vector space? Arch.
Graphics
Theory
NLP
AI
DOCUMENTS AS BAGS OF WORDS DOCUMENT broad tech stock rally may signal trend -­‐ traders. technology stocks rallied on tuesday, with gains scored broadly across many sectors, amid what some traders called a recovery from recent doldrums. INDEX broad may rally rallied signal stock stocks tech technology traders traders trend Doc as vector •  Each doc j is a vector of D×idf values, one component for each term. •  Can normalize to unit length. •  So we have a vector space –  terms are axes -­‐ aka features –  n docs live in this space –  even with stemming, may have 10000+ dimensions –  do we really want to use all terms? Other text categorizaGon applicaGons •  SenGment analysis •  Stylometry SENTIMENT ANALYSIS Id: Abc123 on 5-1-2008 “I bought an iPhone a few days
ago. It is such a nice phone. The touch screen is really
cool. The voice quality is clear too.
It is much better than my old Blackberry, which was a
terrible phone and so difficult to type with its tiny keys.
However, my mother was mad with me as I did not tell
her before I bought the phone. She also thought the
phone was too expensive, …”
SENTIMENT ANALYSIS Id: Abc123 on 5-1-2008 “I bought an iPhone a few days
ago. It is such a nice phone. The touch screen is really
cool. The voice quality is clear too.
It is much better than my old Blackberry, which was a
terrible phone and so difficult to type with its tiny keys.
However, my mother was mad with me as I did not tell
her before I bought the phone. She also thought the
phone was too expensive, …”
SENTIMENT ANALYSIS Id: Abc123 on 5-1-2008 “I bought an iPhone a few days
ago. It is such a nice phone. The touch screen is really
cool. The voice quality is clear too.
It is much better than my old Blackberry, which was a
terrible phone and so difficult to type with its tiny keys.
However, my mother was mad with me as I did not tell
her before I bought the phone. She also thought the
phone was too expensive, …”
HOW DOES IT WORK? •  LEXICON-­‐BASED tools: –  Lookup words in the text in SENTIMENT LEXICON •  E.g., LIWC (Pennebaker) / WordNet Affect / SenGWordnet –  Classify text as posiGve / negaGve if contains a certain number of posiGve / negaGve words •  WORD-­‐BASED tools: –  Use general techniques for supervised text classificaGon to learn which words are best indicators of a parGcular senGment STYLOMETRY Stylometry: Studying properties of the writers of
documents based only on the linguistic style they
exhibit.
!  The best known type of stylometric task: “who wrote
this document?”
! 
! 
! 
“Linguistic Style” Features: sentence length, word
choices, syntactic structure, etc.
Handwriting, content-based features, and contextual
features are not considered.
History of authorship attribution
The classic stylometry problem: The Federalist
Papers.
!  85 anonymous papers to persuade ratification of
the Constitution. 12 of these have disputed
authorship.
!  Stylometry has been used to show Madison
authored the disputed documents.
!  Used as a data set for countless stylometry
studies.
!  Modern Stylometry Based in Machine Learning
!  SVMs, Genetic Algorithms, Neural Networks,
Bayesian Classifiers… used extensively.
! 
Who wrote this?
“On the far side of the river valley
the road passed through a stark
black burn. Charred and limbless
trunks of trees stretching away on
every side. Ash moving over the
road and the sagging hands of blind
wire strung from the blackened
lightpoles whining thinly in the
wind.”
Applications of Stylometry
In the Digital Humanities: Identification of unknown
authors
!  But many other applications as well:
! 
! 
! 
“In some criminal, civil, and security matters, language
can be evidence… When you are faced with a suspicious
document, whether you need to know who wrote it, or if it
is a real threat or real suicide note, or if it is too close for
comfort to some other document, you need reliable,
validated methods.”
Plagiarism, Forensics, Anonymity…
How does it work? Linguistic Features
Basic Measurements:
!  Average syllable/word/sentence count, letter
distribution, punctuation.
!  Lexical Density
!  Unique_Words / Total_Words
!  Gunning-Fog Readability Index:
!  0.4 * ( Average_Sentence_Length +
100 * Complex_Word_Ratio )
!  Result: years of formal education required to read
the text.
! 
Standard stylometric system
! 
Three features: word length, letter usage, punctuation
usage. 95% base accuracy.
Who wrote this?
“On the far side of the river valley
the road passed through a stark
black burn. Charred and limbless
trunks of trees stretching away on
every side. Ash moving over the
road and the sagging hands of blind
wire strung from the blackened
lightpoles whining thinly in the
wind.”
Cormac McCarthy
PRACTICAL CORNER: TOOLS •  As for all other types of classificaGon, can use Weka to learn associaGons between words and authors SENTIMENT ANALYSIS LAB •  In this lab you will see how text categorizaGon using words works in pracGce, in the case of senGment analysis INFORMATION EXTRACTION INFORMATION EXTRACTION •  Goal: being able to answer semanGc queries (a.k.a. “database queries”) using “unstructured” natural language sources •  IdenGfy specific pieces of informaGon in a un-­‐structured or semi-­‐structured textual document. •  Transform this unstructured informaGon into structured relaGons in a database/ontology. SupposiGons: •  A lot of informaGon that could be represented in a structured semanGcally clear format isn’t •  It may be costly, not desired, or not in one’s control (screen scraping) to change this. EXAMPLE OF IE APPLICATION: FINDING JOBS FROM THE WEB foodscience.com-Job2
JobTitle: Ice Cream Guru
Employer: foodscience.com
JobCategory: Travel/Hospitality
JobFunction: Food Services
JobLocation: Upper Midwest
Contact Phone: 800-488-2611
DateExtracted: January 8, 2001
Source: www.foodscience.com/jobs_midwest.htm
OtherCompanyJobs: foodscience.com-Job1
REFERENCES TO (NAMED)
ENTITIES
SITE
LOC
CULTURE
HOW •  Two tasks: –  IdenGfying the part of text that menGons a text (RECOGNITION) –  Classifying it (CLASSIFICATION) •  The two tasks are reduced to a standard classificaGon task by having the system classify WORDS NE: THE IOB REPRESENTATION FEATURES FEATURES FEATURES EVALUATION TYPICAL PERFORMANCE DISAMBIGUATION TO WIKIPEDIA •  Query: •  Wikipedia: Giotto was
called to work in
Padua, and also
in Rimini
81 May 2012 Truc-­‐Vien T. Nguyen OTHER TYPES OF INFORMATION EXTRACTION •  COREFERENCE resoluGon: –  John was late. He should have arrived at 5 … •  RELATION EXTRACTION –  UDO works for the UNIVERSITY OF ESSEX •  CROSS DOCUMENT COREFERENCE TOOLS •  A number of informaGon extracGon tools can be downloaded freely –  NER (standard enGGes): GATE (see lab 1) –  D2W: Wikipedia Miner (see lab 2) PIPELINE AND INFORMATION EXTRACTION LAB •  The lab run by Maha AlthobaiG this aeernoon will explain how to use a standard text mining tool, GATE, to do processing and informaGon extracGon READINGS •  Fabrizio SebasGani. Machine Learning in Automated Text CategorizaGon. ACM Compu<ng Surveys, 34(1):1-­‐47, 2002 •  James Pennebaker. The Secret Life of Pronouns: What Our Words Say About Us. Bloomsbury, 2011. •  Milne & Wi`en (2009). An Open-­‐Source Toolkit for mining Wikipedia.