5 ms·

That seems like quite a natural thing to collect to improve their search algorithm at the very least. "Are people finding what they're looking for?" basically.

by Jemmeh 10y ago

dredmorbius 10y ago
It's not necessary to run trackers on everyone, all the time. Sampling works. Randomly selecting accounts for inclusion and immediately dissociating that account identity from the data would provide substantively useful data. Much the same way as it is possible, on the basis of 100 - 300 randomly selected G+ profiles to gain a strong sense of what the level of public activity of all 2.2 billion profiles (in use as of early 2015) was. Sure, you can increase accuracy, somewhat, by bumping up that sampling count to 50,000 profiles. As I did. Though that (and monte-carlo resampling of subsets of the profiles) substantiated the trends which were clear from the first 100 or so results. Or you can bump up the value to 500,000, as Stone Temple Consulting did. Which again, largely re-substantiated the results I'd found. Google receives something on the order of one billion searches per day. Sampling a very small percentage of those would provide very accurate results, at far smaller processing cost, and moreover, at far smaller data disclosure risk. Data are liability. Even if the concern were, say, edge cases regarding specific terms and the like, that would be possible to explore with targeted random sampling (stratified random samples). All very-well-established statistical techniques. https://ello.co/dredmorbius/post/naya9wqdemiovuvwvoyquq https://ello.co/dredmorbius/post/naya9wqdemiovuvwvoyquq
- Jemmeh 10y ago
  Well just making sure people are finding what they want is only part of it. Since so much of their revenue is from ads, the better their targeting, the more money they make. If they have all this data on a person, they can target on a per person basis. And that money is probably talking louder than any care about liability from data based on how much they're collecting. They're a business, they have to have revenue in some way. I'm making some assumptions, please correct me if I'm wrong.

That seems like quite a natural thing to collect to improve their search algorithm at the very least. "Are people finding what they're looking for?" basically.

by Jemmeh 10y ago

That seems like quite a natural thing to collect to improve their search algorithm at the very least. "Are people finding what they're looking for?" basically.

dredmorbius 10y ago

It's not necessary to run trackers on everyone, all the time. Sampling works. Randomly selecting accounts for inclusion and immediately dissociating that account identity from the data would provide substantively useful data. Much the same way as it is possible, on the basis of 100 - 300 randomly selected G+ profiles to gain a strong sense of what the level of public activity of all 2.2 billion profiles (in use as of early 2015) was. Sure, you can increase accuracy, somewhat, by bumping up that sampling count to 50,000 profiles. As I did. Though that (and monte-carlo resampling of subsets of the profiles) substantiated the trends which were clear from the first 100 or so results. Or you can bump up the value to 500,000, as Stone Temple Consulting did. Which again, largely re-substantiated the results I'd found. Google receives something on the order of one billion searches per day. Sampling a very small percentage of those would provide very accurate results, at far smaller processing cost, and moreover, at far smaller data disclosure risk. Data are liability. Even if the concern were, say, edge cases regarding specific terms and the like, that would be possible to explore with targeted random sampling (stratified random samples). All very-well-established statistical techniques. https://ello.co/dredmorbius/post/naya9wqdemiovuvwvoyquq https://ello.co/dredmorbius/post/naya9wqdemiovuvwvoyquq

Jemmeh 10y ago
Well just making sure people are finding what they want is only part of it. Since so much of their revenue is from ads, the better their targeting, the more money they make. If they have all this data on a person, they can target on a per person basis. And that money is probably talking louder than any care about liability from data based on how much they're collecting. They're a business, they have to have revenue in some way. I'm making some assumptions, please correct me if I'm wrong.