hogwild music!

sequence-based recommendations without needing metadata

Stack & APIs
C++ · Ruby · Redis
Description
== Overview == Rather than use metadata to determine the best fit, we model actual user behavior to determine the optimal playlist. By using the Gracenote API we can use "Moods" to emulate a Songza-like "It's Sunday, I'm feeling like..." interface for users but that's the extent of external metadata. == Technology == Recommendation engine: C++, Redis Web API powered by: Ruby, Neo4j, Redis External APIs (sources of music): Rdio and Gracenote
Comments

Thomas Bouldin

Stripping metadata which was presumably useful for suggest engines and building an engine solely on listening metrics is interesting but short sighted. A listen-based engine is probably simple (i.e. Markov Chain) and has a major grandfathering problem.

JT Bowler

I typed up a ton of feedback and the website died :( But cool idea, just think the data could have been visualized better than a terminal. Also important to make music recommendations using a variety of signals rather than one.

Nick

JT -- Tried writing a long response but the website crashed. The reason for terminal was to show a live example straight from the database using Rdio data.

Nick

The key is to provide recommendations based off of behavior and metadata won't give you sequential order. What you do with other signals is use this to solve the cold-start problem of no listening history of songs.

Nick

Killer product.

Eric Lambrecht

is the the bramo.co UI part of this? The idea is awesome.

Eric Ang

Regarding your list, does the list have to be queued or polled each and every time?

Nick

Which list?

Nick

Which list?

Ashara Ekundayo

an engine that u use to recommend a remix or an original. uses a scoreboard - starts w/ the 1st song and then finds similar songs and finds relationships bewteen artists and users.

Nick

It finds recommendations at the track-level which is more important than artist-level recommendations (what you find most doing)

Karen Wertman

Reccomendation engine. More specific.

Nick

Sequenced-based recommendations. It understands the order of which to recommend data. Let's say you have songs: A, B, C. It understands that you might want to listen Song B after Song A but not listen to Song A after listening to Song B

Nick Quinlan

Rdio integrated at the hackathon Engine, site, etc. not built at the hackathon. http://blamo.co

Nick

I applied the hogwild approach (for the engine) to parallelizing SGD (thus the name) that was done at the hack. LMK and I can send you the paper. Previously, I did not use Hogwild.

Nick

I applied the hogwild approach (for the engine) to parallelizing SGD (thus the name) that was done at the hack. LMK and I can send you the paper. Previously, I did not use Hogwild.

Ching-Wei Chen

Math hack, distance scores. CLI. Unclear

Nick

That data was being pulled directly from the database. The second part of demo shows what the end result becomes.