Posted 23 May 2007 - 05:24 PM
I have used Lycos' Top 50 a lot in the past, and found that the category studies done had some real nuggets in them.
Just visited HotTrends, and I am underwhelmed. It's apparently all machine driven, which could be good, but apparently isnt based on my take today. Guess I will need to study it some more, but the terms i am seeing dont make much sense to me...
1. joya williams
2. monica goodling
3. microwaved baby
4. lucy stoner
5. julianne hough
6. sec baseball tournament
7. rosie and elizabeth
8. fellini film
9. a girl reading painter
10. roman poet
11. school of buddhism
13. jenny heineman
14. singer irene
15. fleet week
16. goddess of wisdom
Maybe I should start watching American Idol or some equally lame cultural pasttime.
Posted 23 May 2007 - 06:22 PM
Consider how they would possibly be calculating "hot trends". Leaving self-fulfilling trends out of the picture (ie people searching for something because other people are searching for it), you might look at the search volume as a curve (approximated -- the absolute values are always "steps, depending on the time-interval they choose to use).
The absolute values of the search volume curve don't say much. Let's say 20% is unique, 60% of the searches are for the top 5% of the terms ("www.google.com", "myspace", etc). That leaves some 20% of the searches in the middle that might be interesting to track. Could you even use those for trends? (not sure)
One way of finding trends is to take the derivative of the search volume curve for a term - this will tell us how fast a term is gaining (or losing) in search volume (slope). An additional factor could be the second derivative of the curve - this will tell us how the "acceleration" for the search term is, how fast is it gaining "slope" (or momentum) [I never had any of this in English, so I might be using the wrong words ; I'm sure I'm getting half of it wrong, so if anyone wants to correct, feel free].
I'm sure Google does more than just some simple math - or at least it does it on millions of items at once (sounds probable), but that would leave us with 3 main factors for any known keyword (and most possibly, for specific geographic areas):
f(x) - the approximated search volume curve: how many searches were done in a certain time period
f'(x) - the first derivative, the rate at which the search volume curve is changing
f''(x) - the second derivative, the rate at with the rate of change for the search volume curve is changing (perhaps I should set up some charts... someday)
Just looking at those three items, for a specific term and a specific geographic region, you could combine them with weights (and more) to determine a "hotness" value for the term (+ region).
so ... why are there so seemingly irrelevant terms listed?
Terms with a low volume have a high change-rate when they fluctuate.
A very, very simplified example:
Term XYZ gets an average of 5 searches/day
Something happens and suddenly XYZ gets 15 searches on that day.
(What are 10 searches more or less? nothing)
The change rate for that would be 200%.
Term ABC gets an average of 500 searches/day
Suddenly it gets 600 searches (+100)
The change rate for that would be 20%
Assuming Google has thresholds in place to prevent the terms with the smallest absolute numbers from being listed (or evalutated), this would still favor small-volume terms over high-volume terms. Who cares if myspace gets a few million searches more today -- if "obscure-term" only gets a few hundred more, it might be seen as a much stronger "trend" or "hot trend".
For simple trends it might be enough to look at the first derivative - but to find hot trends seems to signal that they are looking for the first sign of something taking off -- something they can only do by looking at the second derivative and by mostly ignoring the absolute numbers. Without the absolute numbers being weighted in as much, trends are automatically going to be skewed towards more obscure items.
That said, who knows if Google is doing something as simple as this or if they have added another 100'000 servers to compute some long and twisted formula for each keyword (I personally don't think so).
Where does that bring us? I imagine we'll see a lot of incorrect trend-data there. But it just might show whatever is really going to take off tomorrow in their data as well.
Posted 23 May 2007 - 10:56 PM
They are measuring rate of change and as he points out if you would normally only get three searches for goddess of wisdom and suddenly you got 30000 it will shoot up as the rate of change is increasing rapidly.
As John said:
Without the absolute numbers being weighted in as much, trends are automatically going to be skewed towards more obscure items.
As such they have a high entertainment value Look up for example iGasm! Its a genius little gadget and a marketing lesson. Did you know that Rosie made Elizabeth cry? Or is it the other way round? And what about a girl reading painter? Did this one come from a SEO? Not sure if there are any American idols in there though.
I am pretty much sure like the old keyword tool from overture they can be spammed by automatic searches!
Posted 24 May 2007 - 08:46 AM
I agree with your takes on Google HotTrends.
Soft: Your thoughts on how the algorithm works mathmatically makes a lot sense. I suppose it would need to be different from a list of top search terms in order to be different from what is already out there.
Yannis: Lol of the American Idol. Guess I was sounding a bit dazed and confused there.
I just checked HT today and the top listings are:
1. american idol winner
2. who won american idol
3. american idol finale video
4. working class hero lyrics
5. smokey robinson
6. lost season finale
7. phillippi sparks
8. jordin sparks
9. bette midler
10. gladys knight
11. jimmy conway
12. tony bennett
13. rosie perez
14. wind beneath my wings
15. lost message board
16. joe perry
These micro-tends, now that I think about it more, likely do have some value. But it's not what I was expecting
Edited by Jozian, 24 May 2007 - 08:49 AM.
Posted 24 May 2007 - 09:56 AM
You could however try to analyze the hot trends and try to interpolate what kind of items will be hot trends in the future and prepare by having a site that will cater to those trends ... and of course have your pages listed on top.
But then again, you wouldn't need Google's hot trends for that, just a bit of common sense: ask the people on the street. If you couldn't see "American Idol" coming into the trends then you might want to leave it to someone else .
Posted 24 May 2007 - 10:24 AM
Their won't be searches for google or hotmail.com. No sexual searches. I suspect a whole lot of other searches are filtered as well.
It is a Hot Trends page but of what in where under which parameters... Almost nobody knows
0 user(s) are reading this topic
0 members, 0 guests, 0 anonymous users