3 ms·

Reviews are mathematical garbage even there are real reviewers because we all have different expectations and it varies completely across cultures and geographi

Reviews are mathematical garbage even there are real reviewers because we all have different expectations and it varies completely across cultures and geographies.

lanstin 5y ago
Maybe this signal (fake reviews => fraudulent products) is the most useful info reviews provide.
10000truths 5y ago
Reviews are subjective and qualitative data. Math deals with objective and quantitative data. It’s no surprise that shoehorning the former into the latter is a highly non-trivial problem, that even the best minds in the tech industry struggle to solve for their use cases.
kenniskrag 5y ago
I once watched a movie where the rating was "do you like the item on the left more than on the right". I'm not sure if it is mathematically possible to create a rank from it. I assume, that new items appear and have less comparisons than others.
- kenniskrag 5y ago
  That would remove the bias of what is 5 point of 10 I think.
- nerdponx 5y ago
  This is a technique for "preference elicitation", and related to techniques like Elo scoring and social science fields such as psychometrics. And yes, I think it's much better than reviews that ask for an absolute scale with no context.
- remram 5y ago
  That works when judging aesthetic, but how would that work for extensions though? You can only really judge extensions you have used, and even then how would you choose between your adblocker and your password manager? They do completely different things and I'm not willing to browse without either. edit: I guess the signal "I tried this extension but replaced it with that other one which I like better" would be very informative though
- throwawayboise 5y ago
  I once helped develop a "survey" for a nonprofit org, which wanted to gain some insight on what they were doing well and what they could improve. One of the other people involved kept insisting on reducing the number of questions and complexity of the ratings. He said it all boiled down to one basic question, "would you use this service again" and while we didn't quite get that simple, in retrospect I think he was more right than wrong. Maybe a boolean "would you buy this product again" is the basic question for a review. It's still open to being gamed, but only in one way.

Reviews are mathematical garbage even there are real reviewers because we all have different expectations and it varies completely across cultures and geographies.

lanstin 5y ago

Maybe this signal (fake reviews => fraudulent products) is the most useful info reviews provide.

10000truths 5y ago

Reviews are subjective and qualitative data. Math deals with objective and quantitative data. It’s no surprise that shoehorning the former into the latter is a highly non-trivial problem, that even the best minds in the tech industry struggle to solve for their use cases.

kenniskrag 5y ago

I once watched a movie where the rating was "do you like the item on the left more than on the right". I'm not sure if it is mathematically possible to create a rank from it. I assume, that new items appear and have less comparisons than others.

kenniskrag 5y ago
That would remove the bias of what is 5 point of 10 I think.
nerdponx 5y ago
This is a technique for "preference elicitation", and related to techniques like Elo scoring and social science fields such as psychometrics. And yes, I think it's much better than reviews that ask for an absolute scale with no context.
remram 5y ago
That works when judging aesthetic, but how would that work for extensions though? You can only really judge extensions you have used, and even then how would you choose between your adblocker and your password manager? They do completely different things and I'm not willing to browse without either. edit: I guess the signal "I tried this extension but replaced it with that other one which I like better" would be very informative though
throwawayboise 5y ago
I once helped develop a "survey" for a nonprofit org, which wanted to gain some insight on what they were doing well and what they could improve. One of the other people involved kept insisting on reducing the number of questions and complexity of the ratings. He said it all boiled down to one basic question, "would you use this service again" and while we didn't quite get that simple, in retrospect I think he was more right than wrong. Maybe a boolean "would you buy this product again" is the basic question for a review. It's still open to being gamed, but only in one way.