
BMP
Support
Community
Development
[edit] Picking Songs Automatically[edit] IntroductionThis section details the algorithm to pick songs to play according to the user's taste. [edit] Basic AlgorithmThe user's taste is captured in a series of observations gathered from his or her actions. These actions include overriding an automatic selection. The selection process is a 1storder Markov chain. That is, selections are only based on the song currently playing. Songs played earlier may be taken into account in later revisions of the algorithm. We define a song transition as a change from the song playing at present (hereby, current song), to the song to be selected and played next (hereby, next song). We denote a transition with <math>x \rightarrow y</math>, where <math>x\,</math> is the current song and <math>y\,</math> the next song. Each transition has a probability assigned. The higher its probability, the more likely a transition will be made. The probability of a transition <math>x \rightarrow y</math> is given by <math>P(x \rightarrow y)</math>. The next song to play must always be in the set of tracks available, therefore: <math>\sum_{y \in T}{P(x_0 \rightarrow y)} = 1.0 \qquad x_0 \in T</math> Where <math>T\,</math> is the set of tracks, <math>x_0\,</math> is the current song Since we cannot know what the user likes before any song is played, every transition is equally likely: <math>P(x \rightarrow y) = \frac{1}{T}</math> From then on, <math>P\,</math> is modified according to user actions based on a set of heuristics. [edit] Heuristics

Log in / create account  


This page has been accessed 6,016 times. 