μRaptor: A DOM based system with appetite for hCard elements

LD4IE Challenge 2014 Submission

Transcript

1. μRaptor
   A DOM based system with appetite for hCard elements
   

   View Slide

2. μRaptor
   is hungry
   

   View Slide

3. View Slide

4. Training Phase
   Clean the HTML
   

   View Slide

5. Training Phase
   Clean the HTML
   DOM sub-trees
   

   View Slide

6. Training Phase
   Clean the HTML
   DOM sub-trees
   CSS class co-occurrence
   author
   

   View Slide

7. Training Phase
   Clean the HTML
   DOM sub-trees
   CSS class co-occurrence
   CSS Selectors
   

   View Slide

8. Training Phase
   Clean the HTML
   DOM sub-trees
   CSS class co-occurrence
   Value Constraints
   CSS Selectors
   vcard:email mailto : ALPHA PUNCTUATION ALL_LOWERCASE . ALL_LOWERCASE
   vcard:email mailto : ALPHA PUNCTUATION ALL_LOWERCASE . com
   vcard:email mailto : ALPHA @ ALPHANUMERIC . ALL_LOWERCASE
   vcard:email mailto : ALPHA @ ALPHANUMERIC . com
   vcard:email mailto : ALL_UPPERCASE ****@ ALL_LOWERCASE . ALL_LOWERCASE
   vcard:bday NUMBER - SMALL_NUMBER - SMALL_NUMBER
   vcard:bday MEDIUM_NUMBER - SMALL_NUMBER - SMALL_NUMBER
   We could determine patterns for emails for example:
   … or even for birthdays 
   

   View Slide

9. Extraction Phase
   Clean the HTML
   DOM sub-trees
   CSS class co-occurrence
   Value Constraints
   Pattern Detection
   CSS Selectors
   

   View Slide

10. Extraction Phase
    Clean the HTML
    DOM sub-trees
    CSS class co-occurrence
    Value Constraints
    Pattern Detection
    Elements Qualification
    CSS Selectors
    

    View Slide

11. Clean the HTML
    DOM sub-trees
    CSS class co-occurrence
    Value Constraints
    Pattern Detection
    Elements Qualification
    Models Validation
    CSS Selectors
    Extraction Phase
    RDF Model
    From μRaptor
    RDF Model
    Test set
    ?
    = 0.94
    = 0.7
    = 0.8
    

    View Slide

12. μRaptor
    https://github.com/emir-munoz/uraptor
    

    View Slide

13. We made the discovery of the new μRaptor
    species and I am very pleased some researchers
    helped us understanding its feeding habits
    Godzilla is a doll compared to μRaptor! I am
    currently working on a script for an upcoming
    movie
    As a kid I always wanted to see an actual
    dinosaur. Today my dream comes true
    Damn, he is better than me!
    

    View Slide