Question Answering on the Web

1
Question Answering on the Web

• CMSC 691b
• Rick Carback

2
Presentation Summary

• General Overview
• Architecture of WebQA Systems
• Approaches in WebQA Systems
• Performance
• Proposal
• Testing
• Future Work

3
What is WebQA?

• Making search engines handle questions
• Advantages
• Intuitive
• Better results

4
Traditional QA vs WebQA

• Finding an answer
• Question classification
• IR Engine
• Answer formulation

• Better search results
• Query Formulation
• Search Engine for IR
• Selection and Ranking

5
Problems with the Web

• Large!
• Incorrect Information
• Factual Errors
• Misspellings
• Ambiguity
• Users expect it to be fast

6
General QA Architecture
7
WebQA Architecture
8
General Approaches

• Natural Language Parsing
• Heuristics
• Learning

9
Natural Language Parsing

• Break questions down into parts
• Ranges from deep to shallow
• Used in Query formulation
• Speed is important
• Domain (language) dependent

10
Heuristics

• Direct pattern-matching Rule sets
• Bonuses given for certain conditions
• Language dependent
• Used in query formulation and answer extraction

11
Learning

• Pattern identification and mapping
• Replaces heuristics
• Shallow NLP algorithms
• Language independent

12
Performance

• Answers found 75 of the time without WebQA in
  the first 40 results
• WebQA systems vary in the way they present
  results
• Some Systems use MRR
• Best MRR to date is .46 for these types of systems

13
Proposal

• Learning Approach
• Questions, Answers, and Pages themselves
• More keywords from a first run approach

14
Testing

• Can use TREC system
• Can use 75 finding as a base

15
Future Work

• List questions
• Ambiguous questions
• Multi-lingual focus

16
Questions?