retaining search log data doesn't do much for search quality
It, supposedly, does wonders for targeted advertising.
It also, quite clearly, does wonders for the cosy relationship with 3 letter and 2 letter+number institutions.
Data, it has been argued, is the new oil – the fuel for the information economy – but its importance to search engines may be overstated. In a paper released on Monday through the National Bureau of Economic Research, Lesley Chiou, an associate professor at Occidental College, and Catherine Tucker, a professor at the MIT Sloan …
Targeted advertising is pretty useless - it invariably sends me lots of adverts for something I have already bought, often from the retailer I bought it from. But then I am not one to faff about when I am buying something. YMMV.
I have switched it off on Google now as it was just annoying.
I for one have long had any personalized search results turned off, even if that's not the same as Google not keeping the logs anyway - if nothing else, I simply have a deep aversion for anything that removes consistency and repeatability; my vindication came when a colleague at work simply couldn't locate the same item of interest for a problem we had ("just search for xxxxx, it's right on top" style) no matter what he did and how far he scrolled because he had not turned off personalization and Google deemed he did not need to see that result...
"It is to increase the price of advertising by claiming the adverts only go to people who might buy."
I'm not saying you're wrong but if everyone else sees adverts for stuff they've bought why don't the people who place the ads; and if they do, why don't they realise that this is happening to their ads and that they're paying for junk? Or do the people who place the ads not see them because use ad-blockers like the rest of us because ads are so annoying.
It's not only AI that fails to be self-aware.
There are lot of people in jobs at the White house who owe a lot to the Targetted Advertising campaing that 'won it for Trump' (or well, that's what they think)
OTOH, I think that Amazon are losing the plot on it. I am getting emails from them with book recommendations for books that I bought from them! Perhaps they think that the book wears out?
Posting AC as I don't want even more adverts which TBH are the curse of the modern world.
I think that nails it right there.
You can target your advertising, and there's a big difference in doing it well vs doing it poorly. To use the printer example: I'm shopping for a printer, so I look at lots of printers and accessories (e.g., toner, ink, cables, adapters, whatever). What currently appears to happen is that my search history shows that I've been looking with interest at printers, so it shows me printers. What it should do is log that I've actually bought one, and intelligently use that information to show me ink and such that fits it.
I wonder how much targetted ads would improve if they were curated so that searching for printers shows you ink in the ads. It seems to me that there's an X Y problem here. When you're shopping for a printer, you want to see the X axis, the listing of all the available printers. When you have one, you're shopping on the Y axis, stuff that is only related to that specific printer, but the categorization doesn't work that way.
Exactly. I have my doubts about targeted ads in general, but the utter imbecility of the people programming this stuff is the most obvious [see icon]. If I have just bought a washing machine the balance of probability is that washing machines are the last thing you should be pushing at me. Are the devs utterly clueless or is it just that they don't data mine and classify the product catalogues behind the adverts? Either way, it is embarrassingly stupid.
It looks the algorithm is something along the lines "a random pick in one of the categories items were bought from, especially picks from those items we need to sell".
For example, I've often bought items for a specific camera brand, and never others. Still, the Artificial Idiocy thinks I should buy things for other brands I would never be able to use on my cameras... but it won't show "deals" for my actual brand.
Probably developers have little clues about specific categories needs, and treat everything the same way, with some hilarious results. Probably happens, when you hire cheap ones in poor foreign countries where probably they can't afford most of the items sold.
:If I have just bought a washing machine the balance of probability is that washing machines are the last thing you should be pushing at me.
Therein is the problem. You searched for a washing machine but didn't tell Google you bought one. So what do the algorithms know? I generally block ads so I don't get these "results" and ads. It's annoying as hell but until Google or the advertisers figure out a way to track our purchases and amend their targeting, we'll have to put up with this.
As for Google's privacy.. one needs to login as a user to adjust those settings. No thanks. Why would I want to give them another opening to my PC and browsing?
Well, if it doesn't work using purchasing data, which are much more precise about products you're interested in, and you can afford - how could it work using search data? Whenever I search for "Boeing 737" it's not that I can afford one...
Its said that 'Facebook retains user data for up to 17 months'. But its more like 7+ years and counting. Facebook regularly resurrects Zombie posts 'comments on photos' etc too, as far back as 2010 (account opening). My guess is Facebook need 'public postings' in order to justify user data to advertisers. But since I've posted nothing on my Timeline in 5+ years, Facebook undelete old comments. <Has anyone else seen this>?
BTW: It happens even if you deactivate your account, which I've done for a year. It was an experiment to see if Ad-Data collected under Settings -> Adverts, would continue to pile up from the likes of Datr cookies etc.
I just got a 'memory' from 2006. Facebook keeps everything you've posted, I think what they're talking about there is details like when you logged in and from what IP address. Saving that from 2006 would seem to be of little benefit, but then storage in a cloud the size of Facebook's is essentially infinite so I wonder why they'd bother to delete even that.
Facebook also keeps things such as "likes" around. Does the fact you liked a post 10 years ago really mean anything? Also, at the time, "liking" something was the only way to show you had read and considered a post. Friend had a baby in 2007? Like! Friends parent passed away in 2008? Like!
What useful information does that give Facebook? None. Do you like babies and dead parents, or were you just trying to "support" a friend without having to think up an appropriate sentiment? Are you still friends with these people (real friends, not facebook friends). Facebook can't tell.
A couple of weeks back it was reported that the spy agencies were justifying bulk interception with a quote that they "speak persuasively of developing fragmentary intelligence, of enriching 'seed' information, of following patterns and anomalies, and of the need for the haystack in order to find the needle".
Recent events in London (and repeatedly for over a decade) show that the useless twits already know the terrorists in advance, but seem incapable of doing anything about it in a worrying large percentage of cases. Now the research is in, and it seems that the expectation that more hay produces less needles is in fact the correct one.
Who thinks that the blunderers of the intelligence agencies will now revisit the Snooper's Charter? Nope, I don't either.
"Recent events in London (and repeatedly for over a decade) show that the useless twits already know the terrorists in advance, but seem incapable of doing anything about it in a worrying large percentage of cases. "
Not that I disagree with you in any way - but you only see the ones they couldn't do anything about.
You don't see the people they manage to proactively stop 'doing something'. Furthermore those same agencies are never going to sing about those successes.
"Our successes are secret, you only ever hear about our failures"
To me, success is a conviction based on clear evidence. It is not keeping evidence hidden to identify and ignore a bigger fish or keeping evidence hidden so they can use an explosion to justify a budget increase for more big data.
We have had examples of terrorist killing people despite neighbours supplying clear evidence because the police did not have enough budget to investigate. Cancelling the big data projects and spending the money saved on pay rises and more officers will be more cost effective and there might be enough left over for a clue bat big enough for use on the home secretary.
"To me, success is a conviction based on clear evidence"
Well, to me a success in this field is preventing 'bad thing' from happening.
There are plenty of mechanisms to do that, not all are convictions for the maximum offence.
If we policed the carrying of firearms reasonably strictly then we would expect to have a lower instance of public shootings. I don't define success as convictions for gun murders...
"I don't define success as convictions for gun murders."
It would be a lesser type of success. But wouldn't you agree that a conviction for possession that prevented the murder would be a success?
Initially yes, but in the long term I would define success as even these prosecutions becoming vanishingly rare because the behaviour is eliminated. The obvious measure is reduction in murders...
Note that I don't say elimination, which would of course be ultimate success, but I doubt complete elimination is possible in even this relatively easy case (there are precious few reasons for people to have weapons - there is no reason for a sporting rifle/pistol to be outside of a locked box other than in a range).
The problem is that gray line out there. Here in the States, you have to do something to be arrested and taken off the streets. Being a member, gathering info, gathering "supplies" isn't a crime. The usage is. And as others point out, it's budget. You need to get someone on the inside of the cell, gang, etc. to gather evidence. Costs time and money.
It is a problem at the very basic levels of a free society... What about arresting innocent people? What about locking down civilization such that moment, purchases, even thoughts are scrutinized? Do any of us want that?
"...and of the need for the haystack in order to find the needle".
If i was involved in a tax payer funded organisation, tasked with finding needles, I would expend the resources on facilities that manufacture or sell needles, and then work from there.
I think you'll find it can work remarkably well. It's even shown by respected scientific programmes.
Just watch this when it becomes available again http://www.bbc.co.uk/programmes/b03wcchn You can probably find a knockoff on you tube of it.
Unsurprisingly, it turns out that studies showing the power of placebo all seem to be along the lines of "self-reported pain scale" or similar, and when you really look into them the much more logical conclusion is that placebo (a.k.a. "anything that could obscure the effect you are trying to measure") has exactly the effect that it should. Pill colour makes no difference, sham surgery does not work as well as real surgery (unless the real surgery doesn't work) and the power of the mind cannot cure you.
"I think you'll find it can work remarkably well"
How? Every time a scientific test shows yet again that homeopathy is horseshirt, the credulous saucer loons come out chanting about how it can't be quantified using conventional scientific methods.
Homeopathy is faith-based*, and hence no amout of common sense can be applied to educate those foolish enough to buy into it.
I rather liked the response of a commentator in these hallowed halls who suggested paying for homeopathic medicines with water in which a banknote had been immersed and shaken, then diluted down an appropriate number of times. I felt then as now that this represented wild overpayment, whether the homeopathic dilution effect worked or not.
* faith-based as in "everyone knows it works" rather than "god makes it work". See: Donald Trump supporter.
"How? Every time a scientific test shows yet again..."
Did you look at the link? (Safe to say, I don't believe in Homeopathic medicine)
"...studies showing the power of placebo all seem to be along the lines of..."
Watch the video and many others. The placebo effect is surprisingly powerful and real. Understand what a placebo is and what the placebo effect is. Most scientists would agree about a placebo (and the so-called nocebo effect) effect, it is why drug studies are run double-blind.
The effect is powerful and real, and yes it is why studies are double blinded.
The problem comes when you try to DO anything with this "powerful" effect. This is where you find that studies showing it is possible are not properly blinded, or rely on "self-reported pain scales", or just are poorly designed altogether.
Placebo is an umbrella term for all of the things that can mess with your results in a study. It is not some mystical thing that can be used to treat anything in the real world!
I confess I haven't thought this analogy through much, but I guess it's like saying that a bug in a program did something cool so we should just forget the program and build a solution out of just the bugs!
"It is not some mystical thing that can be used to treat anything in the real world!"
WTF? Had anyone said it was a mystical thing that could be used to treat something. The point was homeopathy isn't real medicine it just has some unfounded results due to the placebo effect.
I might just be making this up or maybe it is something I really remember.
Years ago I saw something trying to justify homoeopathy as 'water memory'. The theory was that due to the way water molecules interact with each other they could 'remember' the shape of the drug molecule and a void was created between the water molecules that was the same shape. Therefore the water exhibited the same properties as the drug (quite a lot of interactions are due to molecular shape) and the treated water was capable of treating the illness. I think they even went as far as trying to use induction coils to try and 'teleport' the molecular shape over electrical wires so that, in theory, you could have your own drugs manufactured at home.
If I have remembered this correctly they were either just making it up or big pharma killed off this idea!
Re: water memory
I think it was even published in Nature or Science (and withdrawn). Still, as they say: there's no nature in Science and no science in Nature...
The placebo effect really is impressive, like with headache medicine working within 10 minutes, when physiologically it takes twice that for the drugs to arrive at their destination. Works even when you know it, which is strange and I dislike that my body is so gullible!
"Did you look at the link? "
Nope. I have better things to do than look for "knock-off youtube links" to content that will suck down an hour of precious mobile bandwidth while I'm commuting in order to understand some point you are trying to make.
If you were making a case against, I already know that.
If you were making a case for, I already ridiculed that.
Moral: If you are going to be ironically clever, don't do it using links to hour long TV shows.
According to Chiou and Tucker, data retention periods didn't affect the flow of traffic from search engines to downstream websites.
Google doesn't really care about downstream websites. What it is interested in is demonstrating the efficacy of its search algorithm to advertisers and it doesn't look like this was covered in the study. To demonstrate the efficacy of the algorithm it constantly needs to check that the results it provides are indeed what users are looking for and constant perusing of the logs is one of the ways to do this, though I suspect aggregation of search terms is main benefit.
I suspect that Google itself knows the value of the data it collects, including the problems associated with ever bigger haystacks. You can see how they do some of these things with Google Trends or the N-Gram research.Trends allow like-for-like comparisons over time, eg. the much touted early indicators of autumn colds and flu. That's gold dust for some but only makes sense with aggregated, anonymised data. I suspect they would probably already pour cold water on the value of ever greater personalisation of adverts, which is where Facebook seems to be going. I think we're going to find out that influence, the holy grail of advertisers, is best reached by the right combination of adverts and content.
On a related note, I am today for the first time turning on Ad Blocking for El. Reg. I regret it - I want to support sites I value and have held off for a long time. But my breaking point is animated ads. You can show as many static ones as you like but if when I am trying to read, an animated bobble head from Hewlett Packard keeps flying up the side of the screen and wobbling at me, I have to block that.
I'm sorry. Just 'No' to auto-play animated and audible ads. It's an instant tab-closer.