music2vec: A smart music recommendation and discovery - PyConSG 2016

Published on: Wednesday, 6 July 2016

Speaker: Laam Pham

Using million playlists as sequences of songs, music2vec (written in Python) can map a song/playlist to a vector. It turns out these vectors make sense and keep the hidden structure. Imagine you like the song A, and dislike song B, but a playlist X. So the math is easy: your findings = song A - song B + playlist X. Voila, this is music2vec.

Music2vec is a recommendation to explore music for you. We combine several state of the art in NLP domain e.g. LDA, word2vec/paragraph2vec. It achieves quite impressive semantic result and can be learnable and smarter day by day. We'll present some basic concepts and hidden technologies behind the scenes for building this application. At the end of the talk, the demo will come out.

Here is the first look and feel "music2vec demo" (

Event Page:

Produced by Engineers.SG

Help us caption & translate this video!