Search Engine Relevance - Computerphile

A search engine can return thousands of web pages, but how does it know whether they are relevant or not? First step is how to measure relevance, as Dr Max Wilson explains.

This video was filmed and edited by Sean Riley.

[mikeydflyingtoaster]

How funny. I went to school with Max! I wonder if he remembers the Dunns. Great to see you doing well!

[the_quad_father]

This is an excellent explanation, but how do search engines determine which results are relevant and not relevant? Only by human evaluation?

[anssiaurum]

If anybody cares at 1:43 it was 69.2%. 70% was surprisingly accurate.

[maxkillers]

U LOST ME WHEN U PUT THOSE MATHS SUMS UP

[wmmcgrailjr]

Fantastic presentation with a simple demonstration training model used when discussing.

[Bluebuthappy]

I like the, "My Lovely Horse" screen saver in the back.

[swamihuman]

SUPER job, Max! Thx.

Now I get to go watch the original two videos, Looking forward to it:)...

[TheJuicyTangerine]

Hey Computerphile, I'd love to see a video on the topic of approximate or "fuzzy" string matching

[damien]

So does it go deeper? If the results are generally the same, and the mean is close... does any evaluation happen which of the them had the most relevant results out of those deemed relevant at all closer to the top?

[koppadasao]

Alan Turing Edition? I wonder…, is the game Turing complete?

[margoumasaleh]

i know i am 7 years late to post this comment and look for an answer but i will just go for it in minute 3:30 you say ("right set of answers ") if i am gonna evaluate my search engine without knowing the right answers can i evaluate my search set

[nO_dNAL]

Nice demonstration but what kind of function is used to actually display the results? I presume it's done through machine learning algorithms where the goal is to maximise the f-score and the average precision.

[kylehazachode]

I still hate the way google broke exact phrase searches. I used quotes all the time for exact phrases. No longer works

[Fionanish]

Any idea how you would build a suitable test case for these queries? i.e is their measure of relevance or will it always be subjective?

[Gullshunter]

Does the search-engine have a self-learning process too? I mean like a retroactive feedback where they give more point to the one that got clicked by the user? (could also be factored by calculating amount of time is spent between the click on the link and the next use of the engine (considering that if it was a bad result the user won't spend much time on it), and the user will look at the other result faster) ?
(could also be factored with many other things too)

[ChrisPowell-TNPEX]

A whole video on relevance without explanation of how you measure relevance per se... And the original requests mentioned Google. I understand the original Page Rank algorithm used max eigenvalues on clusters of incoming references. This is no longer used (or its use is hidden perhaps) because 'link farms' caught on gamifying the algorithm. So Google changed it (and have added a whole bunch of other rules).

But, search engines are not constrained to Web pages. One method to gauge relevancy is where, for example, links with attributes (such as in a graph) between documents are used to measure clustered documents (which implies a relevancy measure). Might have been more relevant (see what I did there?) to take a look at clustering IMHO.

[edgeeffect]

How come there's a picture of Alan Turing on your Monopoly box?

[billyaxlol]

What about the times it takes to get the results?

[Kartaal]

How would you know if a returned result is relevant or not, and more importantly, how would you know how many results that weren't returned are relevant?

In the 5/13 example, how would you figure out the total of relevant results was 13? If the search engine figures out that the remaining 8 are relevant, wouldn't it just return those too?

[TheRealDrWho]

Q. I heard that Bing top searches are google, yahoo... In order to change search engine, is it true???