Live data from Hacker News

Show HN: Similar Phone Finder

productchart.com

11–20 of 29 posts

Re: Show HN: Similar Phone Finder

#11
This looks good. What I was expecting though was a low price phone which match the specs on my phone. I know those phones are out there, but I would like to sort by price (so that I can find cheapest), OS (android/windows/ios), manufacturer.

Re: Show HN: Similar Phone Finder

#13
post #11

This looks good. What I was expecting though was a low price phone which match the specs on my phone. I know those phones are out there, but I would like to sort by price (so that I can find cheapest), OS (android/windows/ios), manufacturer.

Sorting and filtering is pretty much the main functionality of Product Chart. It can be done directly on the smartphone chart:

http://www.productchart.com/smartphones/

The similarity finder is a new addon. I want to find out if "distance based on specs" is an alternative to the usual "customers who bought x also bought y".

Re: Show HN: Similar Phone Finder

#14
Good idea.

However, I was expecting that a Xperia Z1 Compact would be compared with an Xperia Z3 Compact, or at least between similar phones.

It is twice the price and twice the max SD card capacity, but besides that, I think they are quite similar: http://www.productchart.com/smartphones/4597_vs_4674 http://www.phonearena.com/phones/size#/phones/size/Sony-Xper...

Refine the comparing algorithm and it will be awesome.

Re: Show HN: Similar Phone Finder

#19
Your euclidean distance formula is kind of wrong (at least on the site). The units of each term are incompatible so cannot be added together directly, they all should be normalized with a weight factor. A sensible weight factor would be the inverse of the standard deviation of the specific term.

Alternatively you could use the inverse of the covariance matrix as the metric tensor of the properties since there could be quite a few correlations.

Re: Show HN: Similar Phone Finder

#20
post #19

Your euclidean distance formula is kind of wrong (at least on the site). The units of each term are incompatible so cannot be added together directly, they all should be normalized with a weight factor. A sensible weight factor would be the inverse of the standard deviation of the specific term. Alternatively you could use the inverse of the covariance matrix as the metric tensor of the properties since there could b…

The values that go into the formula are normalized, but not weighted. So a similar screen size is "worth" as much as a similar weight. If you find examples, where weighted results would perform better, let me know. You can compare the given phone to other phones by clicking them in the list on the right.
Post reply on HN