Meet Ness, the Pandora for restaurant recommendations
gigaom.com
gigaom.com
It says I have a 84% chance of liking Chipotle... looks like I should give it more data ;)
One of the problems with this I noticed off the bat is that I only gave 4-5 star ratings. Great restaurants stick out in my mind, forgettable meals never. Unless I've had a very negative experience with a restaurant I'm not overly compelled to go and rate it.
[edit] The UI feels really WP7-esque. I like it. It's like a delicious fusion of iOS and WP7... now I'm wondering what the world would be like if we had something like this across the board.
I work with the Machine Learning team at Ness. Thanks for all the feedback! Please email us at support@likeness.com as you have more thoughts, especially about where Ness does and does not work for you.
With regard to only teaching Ness by rating restaurants 4-5 stars, this is a great point, especially for the first 10 ratings. There are a couple particular ways we address this today: 1) Which restaurants you pick tells us a lot. Did you rate hole in the walls vs. expensive places? We have spent a lot of time working on our collaborative filter to address this. Wikipedia has a good article for a starting ground on this, and the papers that came out of the Netflix Prize are great [http://www.netflixprize.com/assets/GrandPrize2009_BPC_BellKo...]. 2) On the personalize page, there are multiple stages. One of them is to confirm that the system is correct about places you don't like, which is obviously also very informative. 3) Sometimes, as with your Chipotle example, we won't be right. Telling us where we are wrong is particularly helpful in correcting the system, which in some sense gives us more information that confirming what we already know. Think of it of correcting a person, who misunderstood what you like from examples you gave. In a metaphorical sense, this is like providing results that have more entropy.
-Jeremy VP of Product & Machine Learning
Thanks for the feedback. We made a slightly different mechanism for rating on the personalize page (the list of places you saw when you first came to the app allowing you to quickly rate). Either a user can skip a place manually by clicking the X, or can scroll down to the bottom and skip all the places that you don't like. You are saying you would rather just scroll past the ones you don't know?
-Jeremy
If the purpose of the Skip All button is to compile a list of restaurants the user has not yet visited, and if this information is sufficiently useful such that you don't want to get rid of the button altogether, perhaps consider a happy medium: for instance, a Skip All of the Above button placed between every ten restaurants on an infinitely scrolling list.
I, for one, think the scroll down -> skip all -> scroll down -> skip all design is clunky--especially since I've visited less than 5% of the restaurants presented at any given time.
-Marissa, Head of Marketing and PR for Ness Computing
Edit: what about a web version or even mobile web version, etc.
I'm a little bummed that I couldn't find some places I've been to on Ness. Maybe their locations DB isn't complete yet.
I found it interesting that I wanted to rate non-food places too, like coworking spaces I've been too. It certainly fits within their "organizing the world's opportunities" goal.
We are working very hard to quickly add all the places in the US, so you can rate the ones you love! Please request these places by clicking the mail button in the bottom right corner of the app, and we will do our best to get them into the system promptly!
We will be building out more channels soon. You mentioned coworking spaces, what else would you like to see?
- Jeremy, VP of Product and Machine Learning at Ness Computing
If you're looking at the use case of locations sought via the mobile app, some suggestions would be: coworking spaces, barber shops/hair salons, groceries stores/supermarkets, running trails, and tailors. At least, as a higher priority.
Lower priority, but still important, would be places that don't need to be that close to me. Meaning, I would rather find a good, reputable service that is far away, than a nearby mediocre service. Such as: dentists, doctors, tax accountants, and mechanics.
I don't think I would need reviews for locations like retail stores, gas stations, and that sort of thing. So that wouldn't be important at all.
This is just a quick swag; I'm sure some of the prioritization could be argued. I hope this helps. Keep up the great work!
But more seriously, this could be a very good app for when you're visiting a new city, and I can think of a dozen people right now who would love to download it but can't...