What is the algorithm for recommendation sites like last.fm, grooveshark, pandora?

Question

I am going to start a project based on a recommendation system. I need to improve myself in this area, which looks like a hot topic on a web page. It is also interesting what is the algorithm lastfm, grooveshark, pandora, which uses recommendations for its system. If you know any book, website or any resource for such algorithms, please inform.

Answer 1

See Collaborative Filtering or Recommended Systems .

One simple Slope One algorithm.

Answer 2

Fashionably late answer: Pandora and Grooveshark are very different in the algorithm they use.

Basically, there are two main approaches to recommendation systems - 1. joint filtering, and 2. based on content. (and hybrid systems)

Most systems are based on collaborative filtering. This basically means matching preference lists): If I liked the elements A, B, C, D, E and F, and some other users liked A, B, C, D, E, F and J - the system will recommend J to me based on the fact that I share the same taste with these users (this is not so simple, but this idea). The main functions that are analyzed here are the identifiers of the elements, and users vote for these elements.

A content-based method analyzes the contents of the elements and creates my profile based on the contents of the elements that I like, and not based on other users.

With that said, Grooveshark is based on collaborative filtering. Pandora is content-based (possibly with one of the co-filtering layers on top).

The interesting thing about Pandora is that the content is analyzed by people (musicians), not automatically. They call it the music genome project ( http://www.pandora.com/mgp.shtml ), where annotators put several marks on each axis on several axes, such as structure, rhythm, tonality, recording technique, and much more (full list : http://en.wikipedia.org/wiki/List_of_Music_Genome_Project_attributes ) This gives them the opportunity to explain and justify the recommended song.

Answer 3

Collective intelligence programming is a nice, affordable introduction to this area.

Answer 4

There is a good demo video with an explanation (and a link to the abstract) on Comparison and visualization of music collections . This approach concerns the analysis of the characteristics of the music itself. Other methods, such as NetFlix and Amazon, are based on recommendations from other users with similar tastes, as well as filtering the main categories.

Answer 5

An excellent article by Yehuda Koren (on the team that won the Netflix prize): BellKor’s Netflix grant (google "GrandPrize2009_BPC_BellKor.pdf" ).

A couple of sites:

• Trustlet.org
• Collaboration tutorials Jun wang
  
• Google: basic element-based recommendation algorithms

Answer 6

Manning also has two good books on this subject. Intelligent Network Algorithms and Collective Intelligence in Action

Answer 7

• Last.fm "neighbors" is probably collaborative filtering.
• Pandora hired hundreds of musicologists to classify songs across 500 dimensions. http://en.wikipedia.org/wiki/Music_Genome_Project

These are two different approaches. Google Scholar is your friend as far as literature is concerned.

Answer 8

The Pandoras algorithm began with the fact that it corresponded to certain genres of music for a specific song. Then it grows slowly by people voting if they like the song or dislike the song, allowing it to eliminate bad songs and promote good songs to the front. He will also be writing new songs that have multiple votes either up or down in your playlist so that the song can get multiple votes.

Not sure if other sites are listed.