Preparing the Text for Indexing 1

Collecting the Documents Tokenizer Token stream. Friends RomansCountrymen Linguistic modules Modified tokens. friend romancountryman Indexer Inverted index. friend roman countryman Documents to be indexed. Friends, Romans, countrymen. 2

What exactly is “a document”? What format is it in? pdf/word/excel/html? What language is it in? What character set is in use? How easy is it to “read it in”? Each of these is a classification problem, that can be tackled with machine learning. But these tasks are often done heuristically … 3

Complications: Format/language What is a unit document? A file? A book? A chapter of a book? An ? An with 5 attachments? A group of files (PPT or LaTeX in HTML) Documents being indexed can include documents from many different languages An index may have to contain terms of several languages. Sometimes a document or its components can contain multiple languages/formats French with a German pdf attachment. 4

Getting the Tokens Tokenizer Token stream. Friends RomansCountrymen Linguistic modules Modified tokens. friend romancountryman Indexer Inverted index. friend roman countryman Documents to be indexed. Friends, Romans, countrymen. 5

6 Words, Terms, Tokens, Types  Word – A delimited string of characters as it appears in the text.  Term – A “normalized” word (case, morphology, spelling etc); an equivalence class of words.  Token – An instance of a word or term occurring in a document.  Type – The same as a term in most cases; an equivalence class of tokens. 6

Tokenization Input: “Friends, Romans and Countrymen” Output: Tokens Friends Romans Countrymen Each such token is now a candidate for an index entry, after further processing But what are valid tokens to emit? 7

What is a valid token? Finland’s capital  Finland? Finlands? Finland’s? Hewlett-Packard  Hewlett and Packard as two tokens? State-of-the-art  break up hyphenated sequence. co-education  ? San Francisco: one token or two? San Francisco-Los Angeles highway: how many tokens? Dr. Summer address is 35 Winter St., , RI, USA. 8

Tokenization: Numbers 3/12/91 Mar. 12, Dec B.C. B-52 My PGP key is 324a3df234cb23e Often, don’t index as text. But often very useful: e.g., for looking up error codes on the web Often, we index “meta-data” separately Creation date, format, etc. 9

Tokenization: language issues East Asian languages (e.g., Chinese and Japanese) have no spaces between words: 莎拉波娃现在居住在美国东南部的佛罗里达。 Not always guaranteed a unique tokenization Semitic languages (Arabic, Hebrew) are written right to left, but certain items (e.g. numbers) written left to right Words are separated, but letter forms within a word form complex ligatures استقلت الجزائر في سنة 1962 بعد 132 عاما من الاحتلال الفرنسي. ‘Algeria achieved its independence in 1962 after 132 years of French occupation.’ 10

Making sense of the Tokens Tokenizer Token stream. Friends RomansCountrymen Linguistic modules Modified tokens. friend romancountryman Indexer Inverted index. friend roman countryman Documents to be indexed. Friends, Romans, countrymen. 11

Linguistic Processing Normalization Capitalization/Case-folding Stop words Stemming Lemmatization 12

Linguistic Processing: Normalization Need to “normalize” terms in indexed text & in query terms into the same form We want to match U.S.A. and USA We most commonly define equivalence classes of terms e.g., by deleting periods in a term Alternative is to do asymmetric expansion: Enter: windowSearch: window, windows Enter: windowsSearch: Windows, windows Enter: WindowsSearch: Windows 13

Normalization: other languages Accents: résumé vs. resume. Most important criterion: How are your users likely to write their queries for these words? Even in languages that standardly have accents, users often may not type them German: Tuebingen vs. Tübingen Should be equivalent 14

Linguistic Processing: Case folding Reduce all letters to lower case exception: upper case (in mid-sentence?)  e.g., General Motors  Fed vs. fed  SAIL vs. sail Often best to lower case everything, since users will use lowercase regardless of ‘correct’ capitalization… 15

Linguistic Processing: Stop Words With a stop list, you exclude from dictionary entirely the commonest words. Why do it: They have little semantic content: the, a, and, to, be They take a lot of space: ~30% of postings for top 30  You will measure this! But the trend is away from doing this: You need them for:  Phrase queries: “King of Denmark”  Various song titles, etc.: “Let it be”, “To be or not to be”  “Relational” queries: “flights to London” 16

Linguistic Processing: Stemming Reduce terms to their “roots” before indexing “Stemming” suggest crude affix chopping language dependent e.g., automate(s), automatic, automation all reduced to automat. Porter’s Algorithm Commonest algorithm for stemming English Results suggest at least as good as other stemming options You find the algorithm and several implementations at

Typical rules in Porter RuleExample sses  sscaresses  caress ies  ibutterflies  butterfli ing  meeting  meet tional  tionintentional  intention Weight of word sensitive rules (m>1) EMENT →  replacement → replac  cement → cement 18

An Example of Stemming After introducing a generic search engine architecture, we examine each engine component in turn. We cover crawling, local Web page storage, indexing, and the use of link analysis for boosting search performance. after introduc a gener search engin architectur, we examin each engin compon in turn. we cover crawl, local web page storag, index, and the us of link analysi for boost search perform. 19

Linguistic Processing: Lemmatization Reduce inflectional/variant forms to base form E.g., am, are, be, is  be car, cars, car's, cars'  car Lemmatization implies doing “proper” reduction to dictionary headword form the boy's cars are different colors  the boy car be different color 20

Phrase queries Want to answer queries such as “stanford university” – as a phrase Thus the sentence “I went to university at Stanford” is not a match. The concept of phrase queries has proven easily understood by users – they use quotes about 10% of web queries are phrase queries In average a query is 2.3 words long. (Is it still the case?) No longer suffices to store only entries 21

Phrase queries via Biword Indices Biwords: Index every consecutive pair of terms in the text as a phrase For example the text “Friends, Romans, Countrymen” would generate the biwords friends romans romans countrymen Each of these biwords is now a dictionary term Two-word phrase query-processing is now immediate (it works exactly like the one term process) 22

Longer phrase queries stanford university palo alto can be broken into the Boolean query on biwords: stanford university AND university palo AND palo alto Without examining each of the returned docs, we cannot verify that the docs matching the above Boolean query do contain the phrase. Can have false positives! (Why?) 23

Issues for biword indexes False positives, as noted before Index blowup due to bigger dictionary For extended biword index, parsing longer queries into conjunctions: E.g., the query tangerine trees, marmalade skies is parsed into tangerine trees AND trees marmalade AND marmalade skies Not standard solution (for all biwords) 24

Better solution: Positional indexes Store, for each term, entries of the form: <number of docs containing term; doc1: position1, position2 … ; doc2: position1, position2 … ; etc.> <be: ; 1: 6 {7, 18, 33, 72, 86, 231}; 2: 2 {3, 149}; 4: 5 {17, 191, 291, 430, 434}; 5: 2 {363, 367}; …> Which of docs 1,2,4,5 could contain “to be or not to be”? 25

Processing a phrase query Merge their doc:position lists to enumerate all positions with “to be or not to be”. to, :  2: 5{1,17,74,222,551}; 4: 5{8,16,190,429,433}; 7: 3{13,23,191};... be, :  1: 2{17,19}; 4: 5{17,191,291,430,434}; 5: 3{14,19,101};... Same general method for proximity searches 26

27 Extra Slides

Combination schemes A positional index expands postings storage substantially (Why?) Biword indexes and positional indexes approaches can be profitably combined For particular phrases (“Michael Jackson”, “Britney Spears”) it is inefficient to keep on merging positional postings lists Even more so for phrases like “The Who” But is useful for “miserable failure” 28

Some Statistics 29 Query2013 results 2013 time 2011 results 2011 time britney spears409,000, ,000, emmy noether930, , the who9,440,000, ,000, wellesley college4,330, , worcester college4,750, , fast cars181,000, ,300, slow cars158,000, , sfjvrbfb wrla10.29

IR: The not-so-exciting part Tokenizer Token stream. Friends RomansCountrymen Linguistic modules Modified tokens. friend romancountryman Indexer Inverted index. friend roman countryman Documents to be indexed. Friends, Romans, countrymen. 30

31 Recap: The Inverted Index For each term t, we store a sorted list of all documents that contain t. 31 DictionaryPostings

32 Intersecting two posting lists 32

Does Google use the Boolean model? The default interpretation of a query [w 1 w 2...w n ] is w 1 AND w 2 AND...AND w n You also get hits that do not contain one of the w i : anchor text page contains variant of w i (morphology, spelling correction, synonym) long queries (n > 10?) when boolean expression generates very few hits  Boolean vs. Ranking of result set  Most well designed Boolean engines rank the Boolean result set  they rank “good hits” higher than “bad hits” (according to some estimator of relevance) 33