Microsoft's new marketing strategy: give up

2013-12-12

I think it’s funny how MS at some point realized they are not the cool kids and there’s no reason to appeal to that target audience. Their new marketing strategy finally admits what’s been long known: the correlation between “business casual” and using Microsoft products:

Bagging as a regularizer

2013-12-06

One thing I encountered today was a trick using bagging as a way to go beyond a point estimate and get an approximation for the full distribution. This can then be used to penalize predictions with larger uncertainty, which helps reducing false positives.

Model benchmarks

2013-11-02

A lot of people have asked me what models we use for recommendations at Spotify so I wanted to share some insights. Here’s benchmarks for some models. Note that we don’t use all of them in production.

statself.com

2013-10-18

Btw I just put something up online that I spent a couple of evenings in my couch putting together: it’s a website where you can track any numerical data on the web. Want to know how many Twitter followers you have? Temperature in NYC? Go to statself.com and start tracking it.

Implicit data and collaborative filtering

2013-09-16

A lot of people these days know about collaborative filtering. It’s that Netflix Prize thing, right? People rate things 1-5 stars and then you have to predict missing ratings.

While there’s no doubt that the Netflix Prize was successful, I think it created an illusion that all recommender systems care about explicit 1-5 ratings and RMSE as the objective. Some people even distrust me when I talk about the approach we take at Spotify.

Vote for our SXSW panel!

2013-09-04

If you have a few minutes, you should check out mine and Chris Johnson‘s panel proposal. Go here and vote: http://panelpicker.sxsw.com/vote/24504

Algorithmic Music Discovery at Spotify

****Spotify crunches hundreds of billions of streams to analyze user’s music taste and provide music recommendations for its users. We will discuss how the algorithms work, how they fit in within the products, what the problems are and where we think music discovery is going. The talk will be quite technical with a focus on the concepts and methods, mainly how we use large scale machine learning, but we will also some aspects of music discovery from a user perspective that greatly influenced the design decisions.

What's up with music recommendations?

2013-08-17

I just answered a Quora question about what, if any, are the differences in the algorithms that are behind recommendations for music and movies.

Of course, every media type is different. For instance, there’s fundamental reasons why latent factor models works really well for music and movies, as opposed to location recommendations where I suspect graph based models are more powerful. People recommendations is another animal and I’m sure beer recommendations has its own domain-specific quirks.

3D

2013-08-12

Andy Sloane decided to call my 2D visualization and raise it to 3D.

(Looks a little weird in the iframe but check out the link). It’s based on a LDA model with 200 topics, so the artists tend to stick to clusters where each cluster is a topic. The embedding also uses t-SNE but in three dimensions (obviously).

2D embedding of 5k artists = WIN

2013-08-11

I’m at KDD in Chicago for a few days. We have a Spotify booth tomorrow, and I wanted to put together some cool graphics to show. I’ve been thinking about doing a 2D embedding of the top artists forever since I read about t-SNE and other papers so this was a perfect opportunity to spend some time on it.