Information Extraction From Automobile Advertisements

1
Information Extraction From Automobile
Advertisements

• Nipun Bhatia
• Rakshit Kumar
• Shashank Senapaty

2
Problem Definition

• Craigslist - Rudimentary keyword search.
• Not a natural way to search for cars.
• Difficult to efficiently find ads with particular
  attributes.
• Want structured search over attributes.
• Attributes Make, Model, Price, Year, Mileage,
  Transmission, PostedBy, Location, Contact

3
Dataset Issues

• 350 postings from the cars trucks section in
  Craigslist.
• Manually annotated with the attributes.

4
System Diagram
5
Feature Selection

• Features
• Title isPresentLexicon, hasDollar_hasDigit,
  hasParanthesis, hasDigit, hasApostrophe_hasDigit,
  PrevLabel, Word
• Body isPresentTrLexicon, isPresentOwLexicon,
  hasDigit_ hasDash, hasDigit_hasDot, hasDigit_
  hasParanthesis, Word_Representation, Neighbor

6
Results
Body Classifier
Title Classifier
7
Example Results
8
Thank You!