Predicting Query Difficulty, Patent granted to IBM Options

[Misscj]

IBM have a really good research team and always publish really interesting papers and file very decent patents.

This one is relevant to SEO because it describes a method for predicting how difficult a query is for a generic search engine.

Patent

"Most search engines have difficulty answering certain queries. For example, consider the query “What impact has the Chunnel had on the British economy and/or the life style of the British?� Most search engines will return many irrelevant documents, containing the words ‘British’, ‘life’, ‘style’, ‘economy’, etc. But the gist of the query, the Chunnel, is usually lost.

There has been a movement to predict the quality of a search results."

How do you think it would be implemented in a search engine?

What do you think it would be most useful for from a user point of view?

How do you think this kind of method would impact current SEO techniques if it was widely implemented?

Reason for edit: Replaced link from page with ads to the real USTPO page

[Nueromancer]

you can imagine the sceen in a few years

"Open the Data centre doors Google bot"

"Ime soory Matt I cant do that"

[Randy]

Well first it's a four year old patent, which is a lifetime in the Internet world.

Second, the type of query given in the example isn't the most difficult for today's search engines to understand. Full Sentence/Full Question types of queries are actually easier than most since the search engine can deconstruct the sentences to sort out what the nouns, verbs, adjectives and adverbs are. This type of sentence deconstruction helps them understand what words are most important.

The much more difficult query types are those that are just single word searches, where the intent of the searcher is much, much harder to decipher. Those are the ones all of the search engines struggle with.

This is why you see those Search Suggestions more often these days than has been the case in the past. It's the search engine's response to generic phrases, where they attempt to suggest closely related phrases that add a bit more context.

[Misscj]

This exactly why I work in conversational agents for IR (IMG:style_emoticons/default/smile.gif)

It's a 4 year patent, but you will find that some things around for quite a long time in the research community don't actually get implemented in public systems immediately. It's old in the internet world, but 4 years in research isn't that long at all!

Agree with your other comments.

[Misscj]

Just wanted to add that for example stemming has been around since 1968, the first stemmer was the Lovins stemmer but Google for example only implemented stemming in something like 2003. I'm sure you know this already, but I just wanted to give an example. It takes ages because some methods don't immediately work with the existing structure of a system. Also some methods are very experimental and not evaluated well enough for a public system to implement them. Even PageRank took 3 years.

It takes a ridiculously long time to figure out which methods go with which combination of other methods.

I'd really like things to go as fast as in the Internet world, because we'd advance really quickly and it'd be exciting, but then it's not possible because it doesn't work the same way. AI for example has been particularly slow (ok it was shut down in the 70's for 10 years - granted). Mathematical logical formalisms take an annoyingly long time.

In 2006 I built something new and the scientists in my field couldn't see the point of it. Now conversational systems are big news, and even though the Turing test has been discounted many times, people still step up to take the challenge, and this area of research is also really slow. The first system was built in 1966 (ELIZA).

I'm not trying to be pedantic, I'm just passionate about these things and wrote a long post by accident. Looks like a blog post, I'll elaborate further there where it's appropriate (IMG:style_emoticons/default/smile.gif)

This post has been edited by Misscj: Oct 21 2008, 04:13 AM