hatkirby / cadence

Twitter bot that listens to music in appropriate situations

cadence

cadence is a Twitter bot that says things about random songs based on their characteristics.

The data that cadence uses to characterize songs comes from the AcousticBrainz project. AcousticBrainz does not currently have an API that would facilitate randomly picking a song, so this bot uses a specially generated datafile that allows it to both randomly choose a song, and identify the type of song it is. Included in this respository is the source of a generator program that can create this datafile from a dump of the AcousticBrainz database.

cadence depends on yaml-cpp to read configuration data from a file, SQLite3 to read a datafile, and my own library libtwitter++ to post to Twitter.

The canonical bot, @songchoicebot, uses the categorization of random songs in order to say things that are ill-fitted to describe the song.

cadence's Issues

Revamp bot

The bot hasn't been updated in a while, and I don't think that the way it currently generates content is the best. It currently chooses a mood for a song based on the mood that AcousticBrainz is most confident about. However, that mood shouldn't exclude the song from being relevant in other moods. The bot should have a table of tracks, and then a table listing the moods for each song, thus allowing a song to have multiple moods. The moods chosen for a track should be those with a confidence over a preset threshold. If a track has no mood with a confidence over the threshold, the song should be assigned a special "unknown" mood. Then, the bot will be able to choose a track uniformly at random, and then choose a random mood applicable to that track. This is preferable to the old behavior, which weighs the probability of a track being picked by the inverse of the number of tracks that have the same mood. It also allows the bot to generate more content per track.