A Machine Learning Approach to Building Domain-Specific Search Engines<br />Presented By:<br />Niharjyoti Sarangi<br />Rol...
Machine Learning<br /><ul><li>  Machine learning is a scientific discipline that is concerned with the design and developm...
   A major focus of machine learning research is to automatically learn to recognize complex patterns and make intelligent...
   A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, ...
General Web search engines :- Attempt to index large portions of the World Wide Web using a Web crawler.
Vertical search engines :- Typically use a focused crawler that attempts to index only Web pages that are relevant to a pr...
  Potential Benefits over general search engines:-</li></ul>Greater precision due to limited scope<br />Leverage domain kn...
Anatomy of a Search Engine<br />Crawling the web<br />Indexing the web<br />Searching the indices<br />Major Data structur...
Web Crawling<br /><ul><li>A Web crawler is a computer program that browses the World Wide Web in a methodical, automated m...
Other terms for Web crawlers are ants, automatic indexers, bots, and worms  or Web spider, Web robot, or—especially in the...
A Web crawler is one type of bot, or software agent. In general, it starts with a list of URLs to visit, called the seeds....
foodscience.com-Job2<br />JobTitle: Ice Cream Guru<br />Employer: foodscience.com<br />JobCategory: Travel/Hospitality<br ...
Information Extraction (contd.)<br />As a task:<br />As a task:<br />Filling slots in a database from sub-segments of text...
Information Extraction (contd.)<br />As a task:<br />Filling slots in a database from sub-segments of text.<br />October 1...
Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br />  segmentation + classif...
Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br />  segmentation + classif...
Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br />  segmentation + classif...
NAME      <br />TITLE   ORGANIZATION<br />Bill Gates<br />CEO<br />Microsoft<br />Bill <br />Veghte<br />VP<br />Microsoft...
Context of Extraction<br />Create ontology<br />Spider<br />Filter by relevance<br />IE<br />Segment<br />Classify<br />As...
IE Techniques<br />Classify Pre-segmentedCandidates<br />Lexicons<br />Sliding Window<br />Abraham Lincoln was born in Ken...
Sliding Window<br />    GRAND CHALLENGES FOR MACHINE LEARNING<br />           Jaime Carbonell<br />       School of Comput...
Sliding Window<br />    GRAND CHALLENGES FOR MACHINE LEARNING<br />           Jaime Carbonell<br />       School of Comput...
Sliding Window<br />    GRAND CHALLENGES FOR MACHINE LEARNING<br />           Jaime Carbonell<br />       School of Comput...
Sliding Window<br />    GRAND CHALLENGES FOR MACHINE LEARNING<br />           Jaime Carbonell<br />       School of Comput...
P(“Wean Hall Rm 5409” = LOCATION) =<br />Prior probabilityof start position<br />Prior probabilityof length<br />Probabili...
Hidden Markov Model<br />HMMs are the standard sequence modeling tool in genomics, music, speech, NLP, …<br />Graphical mo...
IE with HMM<br />Given a sequence of observations:<br />Yesterday Lawrence Saul spoke this example sentence.<br />and a tr...
Limitations of HMM<br />HMM/CRF models have a linearstructure.<br />Web documents have a hierarchicalstructure.<br />
Tree Based Models<br />Extracting from one web site<br />Use site-specificformatting information: e.g., “the JobTitle is a...
Stalker: Hierarchical decomposition of two web sites<br />
Wrapster<br />Common representations for web pages include:<br />a rendered image<br />a DOMtree(tree of HTML markup & tex...
Wrapster<br />html<br />http://wasBang.org/aboutus.html<br />WasBang.com contact info:<br />Currently we have offices in t...
Provo, UT</li></ul>head<br />body<br />…<br />p<br />p<br />“WasBang.com .. info:”<br />ul<br />“Currently..”<br />li<br /...
Upcoming SlideShare
Loading in …5
×

A machine learning approach to building domain specific search

977 views
875 views

Published on

Indexing and crawling the web intelligently by an agent that aids to develop a search engine which goes on learning contextually and semantically.

Published in: Technology, Education
0 Comments
2 Likes
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
977
On SlideShare
0
From Embeds
0
Number of Embeds
17
Actions
Shares
0
Downloads
31
Comments
0
Likes
2
Embeds 0
No embeds

No notes for slide

A machine learning approach to building domain specific search

  1. 1. A Machine Learning Approach to Building Domain-Specific Search Engines<br />Presented By:<br />Niharjyoti Sarangi<br />Roll:06/232<br />8th Semester, B.Tech, IT<br />VSSUT, Burla<br />
  2. 2. Machine Learning<br /><ul><li> Machine learning is a scientific discipline that is concerned with the design and development of algorithms that allow computers to evolve behaviors based on empirical data, such as from sensor data or databases.
  3. 3. A major focus of machine learning research is to automatically learn to recognize complex patterns and make intelligent decisions based on data.
  4. 4. A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.</li></li></ul><li>Vertical Search<br /><ul><li>A vertical search engine, as distinct from a general Web search engine, focuses on a specific segment of online content. The vertical content area may be based on topicality, media type, or genre of content.
  5. 5. General Web search engines :- Attempt to index large portions of the World Wide Web using a Web crawler.
  6. 6. Vertical search engines :- Typically use a focused crawler that attempts to index only Web pages that are relevant to a pre-defined topic or set of topics.</li></li></ul><li>Domain-Specific Search<br /><ul><li>Domain-specific search solutions focus on one area of knowledge, creating customized search experiences, that because of the domain's limited corpus and clear relationships between concepts, provide extremely relevant results for searchers.
  7. 7. Potential Benefits over general search engines:-</li></ul>Greater precision due to limited scope<br />Leverage domain knowledge including taxonomies and ontologies<br />Support specific unique user tasks<br />
  8. 8. Anatomy of a Search Engine<br />Crawling the web<br />Indexing the web<br />Searching the indices<br />Major Data structures<br />Big Files<br />Repositories<br />Document Index<br />Lexicon<br />Hit Lists<br />Forward Index<br />
  9. 9. Web Crawling<br /><ul><li>A Web crawler is a computer program that browses the World Wide Web in a methodical, automated manner or in an orderly fashion.
  10. 10. Other terms for Web crawlers are ants, automatic indexers, bots, and worms or Web spider, Web robot, or—especially in the FOAF community—Web scutter.
  11. 11. A Web crawler is one type of bot, or software agent. In general, it starts with a list of URLs to visit, called the seeds. As the crawler visits these URLs, it identifies all the hyperlinks in the page and adds them to the list of URLs to visit, called the crawl frontier. URLs from the frontier are recursively visited according to a set of policies.</li></li></ul><li>Web Crawling (contd.)<br />
  12. 12. foodscience.com-Job2<br />JobTitle: Ice Cream Guru<br />Employer: foodscience.com<br />JobCategory: Travel/Hospitality<br />JobFunction: Food Services<br />JobLocation: Upper Midwest<br />Contact Phone: 800-488-2611<br />DateExtracted: January 8, 2001<br />Source: www.foodscience.com/jobs_midwest.html<br />OtherCompanyJobs: foodscience.com-Job1<br />Information Extraction<br />
  13. 13. Information Extraction (contd.)<br />As a task:<br />As a task:<br />Filling slots in a database from sub-segments of text.<br />Filling slots in a database from sub-segments of text.<br />October 14, 2002, 4:00 a.m. PT<br />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.<br />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.<br />"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.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />October 14, 2002, 4:00 a.m. PT<br />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.<br />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.<br />"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.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />NAME TITLE ORGANIZATION<br />NAME TITLE ORGANIZATION<br />
  14. 14. Information Extraction (contd.)<br />As a task:<br />Filling slots in a database from sub-segments of text.<br />October 14, 2002, 4:00 a.m. PT<br />For years, Microsoft CorporationCEOBill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation.<br />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.<br />"We can be open source. We love the concept of shared source," said Bill Veghte, a MicrosoftVP. "That's a super-important shift for us in terms of code access.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />IE<br />NAME TITLE ORGANIZATION<br />Bill GatesCEOMicrosoft<br />Bill VeghteVPMicrosoft<br />Richard StallmanfounderFree Soft..<br />
  15. 15. Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br /> segmentation + classification + clustering + association<br />October 14, 2002, 4:00 a.m. PT<br />For years, Microsoft CorporationCEOBill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation.<br />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.<br />"We can be open source. We love the concept of shared source," said Bill Veghte, a MicrosoftVP. "That's a super-important shift for us in terms of code access.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />Microsoft Corporation<br />CEO<br />Bill Gates<br />Microsoft<br />Gates<br />Microsoft<br />Bill Veghte<br />Microsoft<br />VP<br />Richard Stallman<br />founder<br />Free Software Foundation<br />
  16. 16. Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br /> segmentation + classification + association + clustering<br />October 14, 2002, 4:00 a.m. PT<br />For years, Microsoft CorporationCEOBill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation.<br />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.<br />"We can be open source. We love the concept of shared source," said Bill Veghte, a MicrosoftVP. "That's a super-important shift for us in terms of code access.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />Microsoft Corporation<br />CEO<br />Bill Gates<br />Microsoft<br />Gates<br />Microsoft<br />Bill Veghte<br />Microsoft<br />VP<br />Richard Stallman<br />founder<br />Free Software Foundation<br />
  17. 17. Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br /> segmentation + classification+ association + clustering<br />October 14, 2002, 4:00 a.m. PT<br />For years, Microsoft CorporationCEOBill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation.<br />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.<br />"We can be open source. We love the concept of shared source," said Bill Veghte, a MicrosoftVP. "That's a super-important shift for us in terms of code access.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />Microsoft Corporation<br />CEO<br />Bill Gates<br />Microsoft<br />Gates<br />Microsoft<br />Bill Veghte<br />Microsoft<br />VP<br />Richard Stallman<br />founder<br />Free Software Foundation<br />
  18. 18. NAME <br />TITLE ORGANIZATION<br />Bill Gates<br />CEO<br />Microsoft<br />Bill <br />Veghte<br />VP<br />Microsoft<br />Free Soft..<br />Richard <br />Stallman<br />founder<br />Information Extraction (contd.)<br />As a familyof techniques:<br />Information Extraction =<br /> segmentation + classification+ association+ clustering<br />October 14, 2002, 4:00 a.m. PT<br />For years, Microsoft CorporationCEOBill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation.<br />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.<br />"We can be open source. We love the concept of shared source," said Bill Veghte, a MicrosoftVP. "That's a super-important shift for us in terms of code access.“<br />Richard Stallman, founder of the Free Software Foundation, countered saying…<br />Microsoft Corporation<br />CEO<br />Bill Gates<br />Microsoft<br />Gates<br />Microsoft<br />Bill Veghte<br />Microsoft<br />VP<br />Richard Stallman<br />founder<br />Free Software Foundation<br />*<br />*<br />*<br />*<br />
  19. 19. Context of Extraction<br />Create ontology<br />Spider<br />Filter by relevance<br />IE<br />Segment<br />Classify<br />Associate<br />Cluster<br />Database<br />Load DB<br />Query,<br />Search<br />Documentcollection<br />Train extraction models<br />Data mine<br />Label training data<br />
  20. 20. IE Techniques<br />Classify Pre-segmentedCandidates<br />Lexicons<br />Sliding Window<br />Abraham Lincoln was born in Kentucky.<br />Abraham Lincoln was born in Kentucky.<br />Abraham Lincoln was born in Kentucky.<br />member?<br />Classifier<br />Classifier<br />Alabama<br />Alaska<br />…<br />Wisconsin<br />Wyoming<br />which class?<br />which class?<br />Try alternatewindow sizes:<br />Context Free Grammars<br />Finite State Machines<br />Boundary Models<br />Abraham Lincoln was born in Kentucky.<br />Abraham Lincoln was born in Kentucky.<br />Abraham Lincoln was born in Kentucky.<br />Most likely state sequence?<br />NNP<br />V<br />P<br />NP<br />V<br />NNP<br />Most likely parse?<br />Classifier<br />PP<br />which class?<br />VP<br />NP<br />VP<br />BEGIN<br />END<br />BEGIN<br />END<br />S<br />…and beyond<br />Any of these models can be used to capture words, formatting or both.<br />
  21. 21. Sliding Window<br /> GRAND CHALLENGES FOR MACHINE LEARNING<br /> Jaime Carbonell<br /> School of Computer Science<br /> Carnegie Mellon University<br /> 3:30 pm<br /> 7500 Wean Hall<br />Machine learning has evolved from obscurity in the 1970s into a vibrant and popular discipline in artificial intelligence during the 1980s and 1990s. As a result of its success and growth, machine learning is evolving into a collection of related disciplines: inductive concept acquisition, analytic learning in problem solving (e.g. analogy, explanation-based learning), learning theory (e.g. PAC learning), genetic algorithms, connectionist learning, hybrid systems, and so on.<br />CMU UseNet Seminar Announcement<br />
  22. 22. Sliding Window<br /> GRAND CHALLENGES FOR MACHINE LEARNING<br /> Jaime Carbonell<br /> School of Computer Science<br /> Carnegie Mellon University<br /> 3:30 pm<br /> 7500 Wean Hall<br />Machine learning has evolved from obscurity in the 1970s into a vibrant and popular discipline in artificial intelligence during the 1980s and 1990s. As a result of its success and growth, machine learning is evolving into a collection of related disciplines: inductive concept acquisition, analytic learning in problem solving (e.g. analogy, explanation-based learning), learning theory (e.g. PAC learning), genetic algorithms, connectionist learning, hybrid systems, and so on.<br />CMU UseNet Seminar Announcement<br />
  23. 23. Sliding Window<br /> GRAND CHALLENGES FOR MACHINE LEARNING<br /> Jaime Carbonell<br /> School of Computer Science<br /> Carnegie Mellon University<br /> 3:30 pm<br /> 7500 Wean Hall<br />Machine learning has evolved from obscurity in the 1970s into a vibrant and popular discipline in artificial intelligence during the 1980s and 1990s. As a result of its success and growth, machine learning is evolving into a collection of related disciplines: inductive concept acquisition, analytic learning in problem solving (e.g. analogy, explanation-based learning), learning theory (e.g. PAC learning), genetic algorithms, connectionist learning, hybrid systems, and so on.<br />CMU UseNet Seminar Announcement<br />
  24. 24. Sliding Window<br /> GRAND CHALLENGES FOR MACHINE LEARNING<br /> Jaime Carbonell<br /> School of Computer Science<br /> Carnegie Mellon University<br /> 3:30 pm<br /> 7500 Wean Hall<br />Machine learning has evolved from obscurity in the 1970s into a vibrant and popular discipline in artificial intelligence during the 1980s and 1990s. As a result of its success and growth, machine learning is evolving into a collection of related disciplines: inductive concept acquisition, analytic learning in problem solving (e.g. analogy, explanation-based learning), learning theory (e.g. PAC learning), genetic algorithms, connectionist learning, hybrid systems, and so on.<br />CMU UseNet Seminar Announcement<br />
  25. 25. P(“Wean Hall Rm 5409” = LOCATION) =<br />Prior probabilityof start position<br />Prior probabilityof length<br />Probabilityprefix words<br />Probabilitycontents words<br />Probabilitysuffix words<br />Try all start positions and reasonable lengths<br />Estimate these probabilities by (smoothed) counts from labeled training data.<br />If P(“Wean Hall Rm 5409” = LOCATION)is above some threshold, extract it. <br />Naïve Bayes Model<br />00 : pm Place : Wean Hall Rm 5409 Speaker : Sebastian Thrun<br />…<br />w t-m<br />w t-1<br />w t<br />w t+n<br />w t+n+1<br />w t+n+m<br />prefix<br />contents<br />suffix<br />
  26. 26. Hidden Markov Model<br />HMMs are the standard sequence modeling tool in genomics, music, speech, NLP, …<br />Graphical model<br />Finite state model<br />S<br />S<br />S<br />transitions<br />t<br />-<br />1<br />t<br />t+1<br />...<br />...<br />observations<br />...<br />Generates:<br />State<br /> sequence<br />Observation<br /> sequence<br />O<br />O<br />O<br />t<br />t<br />+1<br />-<br />t<br />1<br />o1 o2 o3 o4 o5 o6 o7 o8<br />Parameters: for all states S={s1,s2,…}<br /> Start state probabilities: P(st )<br /> Transition probabilities: P(st|st-1 )<br /> Observation (emission) probabilities: P(ot|st )<br />Training:<br /> Maximize probability of training observations (w/ prior)<br />Usually a multinomial over atomic, fixed alphabet<br />
  27. 27. IE with HMM<br />Given a sequence of observations:<br />Yesterday Lawrence Saul spoke this example sentence.<br />and a trained HMM:<br />Find the most likely state sequence: (Viterbi)<br />YesterdayLawrence Saulspoke this example sentence.<br />Any words said to be generated by the designated “person name”<br />state extract as a person name:<br />Person name: Lawrence Saul<br />
  28. 28. Limitations of HMM<br />HMM/CRF models have a linearstructure.<br />Web documents have a hierarchicalstructure.<br />
  29. 29. Tree Based Models<br />Extracting from one web site<br />Use site-specificformatting information: e.g., “the JobTitle is a bold-faced paragraph in column 2”<br />For large well-structured sites, like parsing a formal language<br />Extracting from many web sites:<br />Need general solutions to entity extraction, grouping into records, etc.<br />Primarily use content information<br />Must deal with a wide range of ways that users present data.<br />Analogous to parsing natural language<br />Problems are complementary:<br />Site-dependent learning can collect training data for a site-independent learner<br />
  30. 30. Stalker: Hierarchical decomposition of two web sites<br />
  31. 31. Wrapster<br />Common representations for web pages include:<br />a rendered image<br />a DOMtree(tree of HTML markup & text)<br />gives some of the power of hierarchical decomposition<br />a sequence of tokens<br />a bag of words, a sequence of characters, a node in a directed graph, . . .<br />Questions: <br />How can we engineer a system to generalize quickly?<br />How can we explorerepresentational choices easily?<br />
  32. 32. Wrapster<br />html<br />http://wasBang.org/aboutus.html<br />WasBang.com contact info:<br />Currently we have offices in two locations:<br /><ul><li>Pittsburgh, PA
  33. 33. Provo, UT</li></ul>head<br />body<br />…<br />p<br />p<br />“WasBang.com .. info:”<br />ul<br />“Currently..”<br />li<br />li<br />a<br />a<br />“Pittsburgh, PA”<br />“Provo, UT”<br />
  34. 34. Wrapster Builders <br /><ul><li>Compose `tagpaths’ and `brackets’
  35. 35. E.g., “extract strings between ‘(‘ and ‘)’ inside a list item inside an unordered list”
  36. 36. Compose `tagpaths’ and language-based extractors
  37. 37. E.g., “extract city names inside the first paragraph”
  38. 38. Extract items based on position inside a rendered table, or properties of the rendered text
  39. 39. E.g., “extract items inside any column headed by text containing the words ‘Job’ and ‘Title’”
  40. 40. E.g. “extract items in boldfaced italics”</li></li></ul><li>Table Based Builders<br />How to represent “links to pages about singers”?<br />Builders can be based on a geometric view of a page.<br />
  41. 41. Wrapster Results<br />F1<br />#examples<br />
  42. 42. References<br />[Bikel et al 1997] Bikel, D.; Miller, S.; Schwartz, R.; and Weischedel, R. Nymble: a high-performance learning name-finder. In Proceedings of ANLP’97, p194-201.<br />[Califf & Mooney 1999], Califf, M.E.; Mooney, R.: Relational Learning of Pattern-Match Rules for Information Extraction, in Proceedings of the Sixteenth National Conference on Artificial Intelligence (AAAI-99).<br />[Cohen, Hurst, Jensen, 2002] Cohen, W.; Hurst, M.; Jensen, L.: A flexible learning system for wrapping tables and lists in HTML documents. Proceedings of The Eleventh International World Wide Web Conference (WWW-2002)<br />[Cohen, Kautz, McAllester 2000] Cohen, W; Kautz, H.; McAllester, D.: Hardening soft information sources. Proceedings of the Sixth International Conference on Knowledge Discovery and Data Mining (KDD-2000).<br />[Cohen, 1998] Cohen, W.: Integration of Heterogeneous Databases Without Common Domains Using Queries Based on Textual Similarity, in Proceedings of ACM SIGMOD-98.<br />[Cohen, 2000a] Cohen, W.: Data Integration using Similarity Joins and a Word-based Information Representation Language, ACM Transactions on Information Systems, 18(3).<br />[Cohen, 2000b] Cohen, W. Automatically Extracting Features for Concept Learning from the Web, Machine Learning: Proceedings of the Seventeeth International Conference (ML-2000).<br />
  43. 43. Niharjyoti Sarangi<br />VSSUT, Burla<br />THANK YOU<br />

×