Thursday, July 16, 2009

Interesting article about semantic search, invariant (and contravariance?)

Ask.com posted an interesting blog article about semantic search. The key idea is to create a set of semantic equivalent phrases and test if the search engine results are invariant to the equivalence.

For instance a query Q1 such as "what is the obama program?" can be rephrased into a query Q2 "can you describe the salient aspects of the Obama program?". Now, Q1 and Q2 are semantically equivalent for humans. Therefore, the expected search results should be invariant (e.g. the same). Unluckly, current search engines do not show such property in many different situations.

I like the idea of invariance, but I suspect that we need to define an auxiliar concept of contravariance. The key intuition is that we can have queries such as "Who is Michelle obama?" and "where is michelle obama?" which are semantically different. Anyway, the returned search results can be the same just because the search engines are stopping the keyword "who" and "where". I believe you may want to consider those negative examples.

No comments:

Post a Comment