Find research datasets worth reusing

Music Streams Labelled with Listening Situation - [User/Track/Device/Situation] Dataset

<p>This is a contextual music dataset labeled with the listening situation associated with each stream.&nbsp; Each stream is composed of the user, track, and device data labelled with a situation. The dataset is collected from Deezer for the period of August 2019 from France and Brazil. The dataset is composed of 3 subsets of&nbsp;situations corresponding to 4, 8, and 12 different situations.&nbsp;&nbsp;The situations are extracted based on keyword matching with the&nbsp;associated playlist title&nbsp;in the Deezer catalog. The full set of situational tags are: &quot;<strong>work, gym, party, sleep | morning, run, night, dance | car, train, relax, club&quot;</strong>.</p> <p>Each instance contains the track/user/deviice triplets, and&nbsp;a situational tag&nbsp;indicating that this&nbsp;user&nbsp;listens to the track in the associated situation wth the corresponding data recieved from the device. The device data contain: &quot;l<strong>inear-time, linear-day, circular-time X, circular-time Y,circular-day X, circular-day Y, device-type, network-type</strong>&quot;.&nbsp;The users are represented as <strong>embeddings</strong> based on their listening history computed through the matrix factorization of the user/track matrix. Additionally, the users are also represented with their demographic data of : &quot;<strong>age, country, gender</strong>&quot;.</p> <p>The creation of the dataset and our experimental results are described in the paper: Karim M. Ibrahim,&nbsp;Elena V. Epure, Geoffroy Peeters,&nbsp;and Ga&euml;l Richard. &quot;Audio Autotagging as Proxy for Contextual MusicRecommendation&quot; [Under Revision].&nbsp;The source code of the paper is available here:&nbsp;<a href="https://github.com/KarimMibrahim/Situational_Session_Generator.git">https://github.com/KarimMibrahim/Situational_Session_Generator.git</a></p> <p>The dataset is composed of the media_id&nbsp;which is the ID of the track in the Deezer catalog.&nbsp;The 30 seconds track previews used to train the model in the paper can be accessed through the Deezer API:&nbsp;<a href="https://developers.deezer.com/api">https://developers.deezer.com/api</a>. Each user is represented with an <strong>anonymized</strong> <strong>user_id</strong> which is associated with the user embedding available in the user_embeddings.npy file. Note: The index of the embeddings in the user_embeddings arrary corresponds to the&nbsp;user_id, i.e.&nbsp;user_id = 100 have its embeddings at&nbsp;&nbsp;user_embeddings[100].&nbsp;</p> <p>Finally, the dataset also contains the&nbsp;splits used in our experiments. Our splits were conditioned by one of three conditions: <em>ColdTrack</em> (no overlap of tracks between the splits), <em>ColdUser</em> (no overlap of users between the splits), and <em>WarmCase</em> (overlaps allowed). Each condition is split into 4 subsets for cross-validation&nbsp;marked with a &quot;<strong>fold</strong>&quot; number in each condition.&nbsp;</p>