Pandora applied to foodNOW

What is the technology behind Pandora? How does Pandora work?

The Music Genome Project

Wikipedia's description: The Music Genome Project, created in January 2000, is an effort founded by Will Glaser, Jon Kraft, and Tim Westergren to "capture the essence of music at the fundamental level" using over 400 attributes to describe songs and a complex mathematical algorithm to organize them. The company Savage Beast Technologies was formed to run the project. A given song is represented by a vector (a list of attributes) containing approximately 150 "genes" (analogous to trait-determining genes for organisms in the field of genetics). Each gene corresponds to a characteristic of the music, for example, gender of lead vocalist, level of distortion on the electric guitar, type of background vocals, etc. Rock and pop songs have 150 genes, rap songs have 350, and jazz songs have approximately 400. Other genres of music, such as world and classical, have 300–500 genes. The system depends on a sufficient number of genes to render useful results. Each gene is assigned a number between 1 and 5, in half-integer increments.

foodNOW

So why can't the same framework be used to describe restaurants? 400 attributes are not necessary, but more than 4 categories (food, service, price and atmosphere) would be necessary to create a useful vector. Comments?

Possible attributes:
-price
-value
-location
-type of food
-type of place
-chef
-trendy
-chain
-take out
-hours
-delivery
-desserts
-liquor diversity
-wine list
-ratings from other sites

This concept can take on many permutations. It would be amazing if the technology could store menus where users could rate their favorite dishes. For example I love carrot cake, but what restaurants in New York serve good carrot cake? Menupages has a function to search by type of food. The website returned a list of restaurants that serve carrot cake, but still ranked via the 4 original categories. Should I assume that because a restaurant is rated well that the place has good carrot cake? No, I cannot come to this conclusion without reading through countless reviews that may or may not talk about dessert. So I what would like what my application to allow you to put in a specific food where it then would provide suggestions on where to find it!

My big idea could offer users an accurate cross-content and cross platform search by taste engine.

Bloomberg Function Comparision

I found a version of my idea at work. Another parallel can be drawn to a function that I use in Bloomberg called COMB. In finance we use COMB to determine a bond's relative value as compared to other bonds with similar maturities, ratings, currencies, and industry classifications. COMB use a default search to find comparable bonds and displays the current price, yield, and spreads for each security.

So basically in English this means when I have a bond load into the system I can then ask Bloomberg to search though thousands of other bonds to find a similar bond. This function is like a thesaurus for bonds. Note, one does not have to modify the filter for the search to be successful. Bonds are more like words, while restaurants have more song like qualities, ie. there is some qualitative aspect that needs interpretation.

How does a computer predict behavioral patterns?

I was speaking to my father about my idea this morning and he was telling me how he can relate to this in his field, science. He said, "In medical science we do computer modeling. We have a bunch of laboratory data that is associated with a clinical out come. So we collect samples from a statistically significant number of patients usually in the hundreds to thousands and train the computer to learn that a laboratory parameter is associated with a clinical outcome. Then we collect another set of samples usually in ten to hundred thousands and actually do the analysis asking whether that lab parameter is associated with that clinical outcome. You collect enough preliminary preferences through his/her usage and the computer learns his/her preferences. You do this for thousands and thousands of users and you can statistically predict that an individual will select a certain choice. Thus you have come up with a formula for predicting success."

Although this is an interesting concept, collecting enough data so you can predict a future pattern, it is different than my initial concept for foodNOW. foodNOW should already have a "formula" for predicting whether a person would like a restaurant based on his/her comparable choice. Of course this "formula" or technology should by dynamic and take user preferences to improve upon future suggestions.

Naive Bayes Classifier

I was told that I should explore Bayesian statistics in trying to further understand how a computer can learn. This seems like another way to classify or code a data set.

My go to site Wikipedia explains, "A naive Bayes classifier is a term in Bayesian statistics dealing with a simple probabilistic classifier based on applying Bayes' theorem with strong (naive) independence assumptions. A more descriptive term for the underlying probability model would be "independent feature model".

In simple terms, a naive Bayes classifier assumes that the presence (or absence) of a particular feature of a class is unrelated to the presence (or absence) of any other feature. For example, a fruit may be considered to be an apple if it is red, round, and about 4" in diameter. Even though these features depend on the existence of the other features, a naive Bayes classifier considers all of these properties to independently contribute to the probability that this fruit is an apple.

Depending on the precise nature of the probability model, naive Bayes classifiers can be trained very efficiently in a supervised learning setting. "

Ingenious Genius Feature

Apple has a new feature in iTunes which is an instant playlist. Elliot Feldman of the Associate Press explains, "Genius is a feature of iTunes 8 that creates instant playlists from the music that's in your iTunes Library. Just select a song to play from your iTunes Library, click on the Genius button (an atomic symbol) on the bottom row of the iTunes 8 browser interface, and an instant playlist of songs is generated from your iTunes Library. Along with the instant playlist generated from your iTunes Library, Genius also generates a sidebar list of compatible songs that you can purchase from the online iTunes Store."

Now how does this work? I was told that they cross-reference data from everyone's playlists so if a lot of people like Madona and they also like Britney Spears then iTune's uses this data to make the Genius playlist. A verision of Bayesian statistics at work here? Now this might be the cheaper and only way data about restaurants can be gathered because unlike songs, restaurants constantly change. So is a "restaurant genome project" not possible? Just something to think about here.

Google Maps

I downloaded google maps the other day and found out about a technology called my latitude. It allows you to "See your friends' locations and status messages and share yours with them." This could be considered a invasion of privacy, but indeed it is a useful tool. It apparently uses cell phone towers to approximate your location. Google maps gave my cell phone GPS functions without having GPS. Now, I wonder if it is possible for foodNOW to be linked to or connect to the google maps function or have this function embedded in the application itself. This might be asking too much of one little application.

Mobile Social Networking Users

eMarketer forecasts that mobile social networking will grow from 82 million users in 2007 to over 800 million worldwide by 2012.

Clearly this proves that the user base is there and growing quickly. If the correct application is developed I think I could be well positioned to capitalize off this growth in mobile social networking.













The target market is available and only growing!

Naming...

I have not given much thought to the name of this technology, my "Restaurant Genome Project".

-Foodengineer
-Foodengine
-Food + Predict
-Food + Guess
-Food for thought.
-iChow (there is already mychow on chowhound)
-foodNOW (this might be the best so far?)
-FeedMe
-Food + Thesaurus = foodasaurus

Big Idea!

One day this idea just hit me and I called up my friend Andy to tell him that I thought of a great idea. He is my go to entrepreneur friend that would appreciate something like this. My great idea was Pandora except for restaurants. Initially I thought this would be a good website, but there are already so many out there in this space--just to name a few Citysearch, Menupages and Urban Spoon. It is worth mentioning that none of those sites are really are predictive in nature. I am not sure how Pandora works, but if this could be done with restaurants and not songs, I know the application (whether a website or mobile phone application)would be successful because it solves a common problem that everyone has.

As we conversed more, the idea got better. I wanted to develop an iPhone application that would combine the Urban Spoon iPhone app + Pandora.