Hacker News Reader

Comment by adtac | Hacker News Reader

>We moderate every report, with both a back end review, and front end/human review of each submission with the goal of eliminating malicious and inauthentic reports

Humans don't scale. How do you plan on reliably doing this when you get larger?

>With our data, we predicted that Chipotle would have food safety issues ahead of their series of outbreaks, and we detected several of their outbreaks, and many others in real time.

Not to sound pessimistic, but I'd also like to know if you had any false flags? Where you predicted an outbreak and it never happened? As you can imagine, falsely publishing something like this could be hugely detrimental to a company's reputation.

Cool idea, just curious about certain aspects of your operation :)

View on HN

Thanks!

Scale > Great question. We think we can do it. We have a great offshore team with affordable scaling, also we just implemented our first machine learning module, and plan to build on this to help speed up and improve accuracy.

False flags on detecting outbreaks > No we have not. Mainly becasue we don't make pronouncements such as "Restaurant X made people sick and they have a problem". Mainly because foodborne illness can be complicated. It may have been a supplier problem, it may be something related to that restaurant but not that restaurants fault. What our site does make clear is that 120 people reported sick after eating at Restaurant X, and at that point both the restaurant and public health ought to examine the situation immediately. It's up to public health agencies, the epidemiologists etc. to make the official judgement.

False flags on predicting food safety issues (potential precursor to outbreaks) > So far our goal is to partner with companies that exhibit high risk and the health agencies that monitor them. So we have not made any public predictions on that, except I think once on Chipotle, which was correct, but I'm not certain we would do that often. This will probably evolve, and it's something we need to be careful on.

A single human doesn't scale, a work force of multiple humans can.

With something this sensitive I would be very wary of machine classification.

Agreed, I expect we would use it to supplement but not supplant human reviews.

More importantly, what's stopping a competitor from submitting a bunch of fake reports? The authenticity isn't inherent from the text of the report. Adding friction won't help here.

Agreed. The text of the report is only a small part of it. We are looking at a range of other things on the back end. Also a scenario like that is a red flag to us, and it triggers further due diligence. It'will also trigger public health, who will then require people to pick up the phone, it will require a 72 hour food history so on, physical site investigation. It does not work like 50 reports come in, and an outbreak is immediately declared - far from it. That is just the start of further work. Not to say someday the system won't be gamed, but we are doing a lot of hard work to ensure it does not happen.

You throw around high report numbers, but most listed locations have one report.

I would probably like it to receive a warning about a recent report a place has while I'm there.

How about user verification, perhaps even with a verified doctor visit getting a higher trustworthiness rating?

I would consider DineSafe to be a nicer, more memorable, internationally better-faring name than iwaspoisoned.

We roll data off the site after 30 days, so you are only looking at recent history, and also not every report goes live on the site. I think the largest we had reported in this period are maybe a couple of cases citing between 20 and 40 people sick.

Thanks for the sentiment re Dinesafe. Iwaspoisoned.com is not for everyone, but it does fare well on search and being memorable.

Reply on news.ycombinator.com