1. 1 I256: Applied Natural Language Processing Preslav Nakov and Marti Hearst October 16, 2006 (Many slides originally by Barbara Rosario, modified here)

2. 2 Today Classification Text categorization (and other applications) Various issues regarding classification Clustering vs. classification, binary vs. multi-way, flat vs. hierarchical classification… Introduce the steps necessary for a classification task Define classes Label text Features Training and evaluation of a classifier

3. 3 From: Foundations of Statistical Natural Language Processing. Manning and Schutze Classification Goal: Assign ‘objects’ from a universe to two or more classes or categories Examples: Problem Object Categories Tagging Word POS Sense Disambiguation Word The word’s senses Information retrieval Document Relevant/not relevant Sentiment classification Document Positive/negative Author identification Document Authors

4. 4 Slide adapted from Paul Bennet Text Categorization Applications Web pages organized into category hierarchies Journal articles indexed by subject categories (e.g., the Library of Congress, MEDLINE, etc.) Responses to Census Bureau occupations Patents archived using International Patent Classification Patient records coded using international insurance categories E-mail message filtering News events tracked and filtered by topics Spam vs. anti-palm

5. 5 Yahoo News Categories

6. 6 Slide adapted froml Paul Bennett Why not a semi-automatic text categorization tool? Humans can encode knowledge of what constitutes membership in a category. This encoding can then be automatically applied by a machine to categorize new examples. For example...

7. 7 Slide adapted froml Paul Bennett Expert System (late 1980s)

8. 8 Slide adapted froml Paul Bennett Rule-based Approach to Text Categorization Text in a Web Page “Saeco revolutionized espresso brewing a decade ago by introducing Saeco SuperAutomatic machines, which go from bean to coffee at the touch of a button. The all-new Saeco Vienna Super-Automatic home coffee and cappucino machine combines top quality with low price!” Rules Rule 1. (espresso or coffee or cappucino ) and machine* Coffee Maker Rule 2. automat* and answering and machine* Phone Rule...

9. 9 Slide adapted froml Paul Bennett Defining Rules By Hand This is fine for low-stakes applications Google and Yahoo alerts allow users to automatically receive news articles containing certain keywords Called “filtering” or “routing” Works fine when it’s ok to miss some things But when high accuracy is required, experience has shown too time consuming too difficult inconsistency issues (as the rule set gets large)

10. 10 Slide adapted froml Paul Bennett Replace Knowledge Engineering with a Statistical Learner

11. 11 Slide adapted froml Paul Bennett Cost of Manual Text Categorization Yahoo!  200 (?) people for manual labeling of Web pages  using a hierarchy of 500,000 categories MEDLINE (National Library of Medicine)  $2 million/year for manual indexing of journal articles  using MEdical Subject Headings (18,000 categories) Mayo Clinic  $1.4 million annually for coding patient-record events  using the International Classification of Diseases (ICD) for billing insurance companies US Census Bureau decennial census (1990: 22 million responses)  232 industry categories and 504 occupation categories  $15 million if fully done by hand

12. 12 Slide adapted froml Paul Bennett Knowledge Statistical Engineering Learning For US Census Bureau Decennial Census 1990 232 industry categories and 504 occupation categories $15 million if fully done by hand Define classification rules manually: Expert System AIOCS Development time: 192 person-months (2 people, 8 years) Accuracy = 47% Learn classification function Nearest Neighbor classification (Creecy ’92: 1-NN) Development time: 4 person-months (Thinking Machine) Accuracy = 60% vs.

13. 13 Text Topic categorization Topic categorization: classify the document into semantics topics The U.S. swept into the Davis Cup final on Saturday when twins Bob and Mike Bryan defeated Belarus's Max Mirnyi and Vladimir Voltchkov to give the Americans an unsurmountable 3-0 lead in the best-of-five semi-final tie. One of the strangest, most relentless hurricane seasons on record reached new bizarre heights yesterday as the plodding approach of Hurricane Jeanne prompted evacuation orders for hundreds of thousands of Floridians and high wind warnings that stretched 350 miles from the swamp towns south of Miami to the historic city of St. Augustine.

14. 14 The Reuters collection A gold standard Collection of (21,578) newswire documents. For research purposes: a standard text collection to compare systems and algorithms 135 valid topics categories

15. 15 Reuters Top topics in Reuters

16. 16 Reuters Document Example 2-MAR-1987 16:51:43.42 livestock hog AMERICAN PORK CONGRESS KICKS OFF TOMORROW CHICAGO, March 2 - The American Pork Congress kicks off tomorrow, March 3, in Indianapolis with 160 of the nations pork producers from 44 member states determining industry positions on a number of issues, according to the National Pork Producers Council, NPPC. Delegates to the three day Congress will be considering 26 resolutions concerning various issues, including the future direction of farm policy and the tax law as it applies to the agriculture sector. The delegates will also debate whether to endorse concepts of a national PRV (pseudorabies virus) control and eradication program, the NPPC said. A large trade show, in conjunction with the congress, will feature the latest in technology in all areas of the industry, the NPPC added. Reuter

17. 17 Classification vs. Clustering Classification assumes labeled data: we know how many classes there are and we have examples for each class (labeled data). Classification is supervised In Clustering we don’t have labeled data; we just assume that there is a natural division in the data and we may not know how many divisions (clusters) there are Clustering is unsupervised

18. 18 Classification Class1 Class2

19. 19 Classification Class1 Class2

20. 20 Classification Class1 Class2

21. 21 Classification Class1 Class2

22. 22 Clustering

23. 23 Clustering

24. 24 Clustering

25. 25 Clustering

26. 26 Clustering

27. 27 Categories (Labels, Classes) Labeling data 2 problems: Decide the possible classes (which ones, how many) Domain and application dependent Label text Difficult, time consuming, inconsistency between annotators

28. 28 Reuters Example, revisited 2-MAR-1987 16:51:43.42 livestock hog AMERICAN PORK CONGRESS KICKS OFF TOMORROW CHICAGO, March 2 - The American Pork Congress kicks off tomorrow, March 3, in Indianapolis with 160 of the nations pork producers from 44 member states determining industry positions on a number of issues, according to the National Pork Producers Council, NPPC. Delegates to the three day Congress will be considering 26 resolutions concerning various issues, including the future direction of farm policy and the tax law as it applies to the agriculture sector. The delegates will also debate whether to endorse concepts of a national PRV (pseudorabies virus) control and eradication program, the NPPC said. A large trade show, in conjunction with the congress, will feature the latest in technology in all areas of the industry, the NPPC added. Reuter Why not topic = policy ?

29. 29 Binary vs. multi-way classification Binary classification: two classes Multi-way classification: more than two classes Sometime it can be convenient to treat a multi-way problem like a binary one: one class versus all the others, for all classes

30. 30 Flat vs. Hierarchical classification Flat classification: relations between the classes undetermined Hierarchical classification: hierarchy where each node is the sub-class of its parent’s node

31. 31 Single- vs. multi-category classification In single-category text classification each text belongs to exactly one category In multi-category text classification, each text can have zero or more categories

32. 32 Features >>> text = "Seven-time Formula One champion Michael Schumacher took on the Shanghai circuit Saturday in qualifying for the first Chinese Grand Prix." >>> label = “sport” >>> labeled_text = LabeledText(text, label) Here the classification takes as input the whole string What’s the problem with that? What are the features that could be useful for this example?

33. 33 Feature terminology Feature: An aspect of the text that is relevant to the task Some typical features Words present in text Frequency of words Capitalization Are there NE? WordNet Others?

34. 34 Feature terminology Feature: An aspect of the text that is relevant to the task Feature value: the realization of the feature in the text Words present in text : Kerry, Schumacher, China… Frequency of word: Kerry(10), Schumacher(1)… Are there dates? Yes/no Are there PERSONS? Yes/no Are there ORGANIZATIONS? Yes/no WordNet: Holonyms (China is part of Asia), Synonyms(China, People's Republic of China, mainland China)

35. 35 Feature Types Boolean (or Binary) Features Features that generate boolean (binary) values. Boolean features are the simplest and the most common type of feature. f 1 (text) = 1 if text contain “Kerry” 0 otherwise f 2 (text) = 1 if text contain PERSON 0 otherwise

36. 36 Feature Types Integer Features Features that generate integer values. Integer features can be used to give classifiers access to more precise information about the text. f 1 (text) = Number of times text contains “Kerry” f 2 (text) = Number of times text contains PERSON

37. 37 Feature selection How do we choose the “right” features? A future lecture

38. 38 Classification Define classes Label text Extract Features Choose a classifier >>> my_classifier.classify(token) The Naive Bayes Classifier NN (perceptron) SVM …. Train it (and test it) Use it to classify new examples

39. 39 Training Usually the classifier is defined by a set of parameters Training is the procedure for finding a “good” set of parameters Goodness is determined by an optimization criterion such as misclassification rate Some classifiers are guaranteed to find the optimal set of parameters

40. 40 Testing, evaluation of the classifier After choosing the parameters of the classifiers (i.e. after training it) we need to test how well it’s doing on a test set (not included in the training set) Calculate misclassification on the test set

41. 41 Evaluating classifiers Contingency table for the evaluation of a binary classifier GREEN is correctRED is correct GREEN was assignedab RED was assignedcd Accuracy = (a+d)/(a+b+c+d) Precision: P_GREEN = a/(a+b), P_ RED = d/(c+d) Recall: R_GREEN = a/(a+c), R_ RED = d/(b+d)

42. 42 *From: Improving the Performance of Naive Bayes for Text Classification, Shen and Yang Training size The more the better! (usually) Results for text classification *

43. 43 *From: Improving the Performance of Naive Bayes for Text Classification, Shen and Yang Training size

44. 44 *From: Improving the Performance of Naive Bayes for Text Classification, Shen and Yang Training size

45. 45 Authorship Attribution a Comparison Of Three Methods, Matthew Care Training Size Author identification

46. 46 Upcoming Classifiers Feature selection algorithms