SenseAble Search

1
SenseAble Search

• Shailesh Kochhar and Adam Vogel
• CS 498CXZ, Spring 2006

2
Ambiguity

• Words are ambiguous
• But they dont have to be
• Otherwise wed never understand each other

3

• So now, when is

4

• A golf club not a golf club?
• A chair not a chair?
• A bill not a bill?

5

• When

6

• It allows you to rent a golf cart.
• He or she calls a meeting to order.
• People vote on it, or it comes with a duck

7
Ambiguity affects search
8

• Sometimes we dont know what a query means

9
Muckey Mouse?
10

• Can we do the same for meaning?

11
Ambiguity Detection

• Tag sense of query terms in top documents
• Examine the distribution of senses
• Ambiguity Large number of senses
• Diverse distribution More random
• Measure randomness?

12
Ambiguity Resolution

• Use entropy of the top n results
• Set a threshold for the entropy
• Pick most likely senses
• Ask Did you mean ?

13
Ranking with Sense

• Simple filtering
• More complex
• Term-sense frequency
• IDF with respect to (term, sense)

14

• What about extremely rare senses?

15
Sense Diversification

• Sense of top results vs. all relevant docs
• If difference is large, suggest rare senses to
  user

16
Disambiguation Observations

• WordNet senses are fine-grained
• Small DA noise gt large entropy noise
• Short queries gt ambiguity

17
Demo

• http//csil-linux40.cs.uiuc.edu8080/