Comment by mNovak | Hacker News Reader

Comment by mNovak | Hacker News Reader

I like to buy used electronics online, but it's very tedious cross-referencing model numbers and benchmarks and such to determine what's actually a good deal. For example which is better, a 10th gen i5 or a 7th gen i7 at the same price?

So as a pandemic hobby, I made a website to scrape eBay laptop listings, extract all the relevant data, cross reference against benchmarks, and sort the result by "value" (aggregate performance / cost). I took inspiration from the parametric search on site like digikey for commodity components, where no one cares about brand.

You can check it out here [1]. Maybe eventually I'll write a showHN, but I haven't been working on it for a while, and there's a few features I want to add before that. Originally I planned to use affiliate links, which seemed like a nice all-parties-win setup, but eBay kicked me off the Partner Network program for not generating traffic. So presently it's totally free and non-profitable.

[1] https://bytebucket.co

This is awesome! I regularly encounter the same problem. Particularly when shopping for electronics, where there is a clear metric that counts (say, USD/Watt hour for batteries). But how often had I wished something like this existed for other items as well. Say, toothpaste. What do I care about the brand? I care about the abrasivity factor (RDA), and the amount of flouride. Or sweaters. I don't care that it looks like wool, or that other people use it as replacement for wool. Instead, I want something that's 100 % wool, and no acrylics. Or think supplements. I don't care about fancy pictures of healthy people. I want to know what's in the capsule, where it was source, and how much of it I get per USD.

I get that a site like yours cannot solve all of that. You can only extract the info that's already there. And parsing such data for different categories of products is a major league hassle.

I once built a tool that did this for used cars. And it would allow you to plot all available listings that matched your filters in a price/milage scatter graph. To make the outliers really visible.

Can you share how you solved the parsing problem in your implementation? Is it just regexes, or are you employing NLP span categorization techniques?

It's just a bunch of regex, nothing too sophisticated. Generally the people making listings are trying to convey all the data, so in a sense it's semi-structured. The most limiting factor is getting a CPU model; I'd like to eventually build up a taxonomy of major device brands/models to map to their specs (I do have this currently for Apple, which many people post year or A-model numbers for).

Reply on news.ycombinator.com