Artificial intelligence based music playlist reordering and song performance assessment

A method for artificial intelligence based music playlist reordering and song performance assessment, the method comprising:

• ascertaining, by at least one hardware processor, listening data for a plurality of tracks;
• generating, by the at least one hardware processor and based on an analysis of the listening data, a plurality of sessions;
• generating, by the at least one hardware processor and based on the plurality of sessions, a user-song interaction graph by
• generating, for each session of the plurality of sessions, a linear graph based on a sequence of tracks played, and
• superimposing each linear graph on each other linear graph based on edges present between nodes, wherein a node represents a unique track played, an edge represents a link to a next track that is played, and a weight of an edge represents a number of times the edge is present across the plurality of sessions;
• obtaining, by the at least one hardware processor, an input playlist;
• utilizing, by the at least one hardware processor, the user-song interaction graph to generate, based on the input playlist, a modified playlist;
• determining, by the at least one hardware processor and for each track of the input playlist, a session user density based on a pagerank value of a track based on the user-song interaction graph, wherein the pagerank value of the track represents a total number of listening events of the track;
• determining, by the at least one hardware processor and for each track of the input playlist, a niche factor as a ratio of a total number of listening events of a track to a number of unique listeners of the input playlist;
• determining, by the at least one hardware processor and for each track of the input playlist, a song skip probability as a probability that a track will be skipped in the input playlist;
• determining, by the at least one hardware processor and for each track of the input playlist, a composite score as a function of the session user density, the niche factor, and the song skip probability;
• and
• identifying, by the at least one hardware processor, based on the composite score and based on a heuristic based path finding model, a path in the user-song interaction graph that includes a maximum user retention path probability.

A non-transitory computer readable medium having stored thereon machine readable instructions, the machine readable instructions, when executed by at least one hardware processor, cause the at least one hardware processor to:

• ascertain listening data for a plurality of tracks;
• generate, based on an analysis of the listening data, a plurality of sessions by
• determining, for each track played in the listening data, a user identification, a track identification, and a timestamp associated with playing of a track,
• generating groups from the listening data according to the user identification and the timestamp,
• filtering the generated groups to determine groups that include a minimum count of unique tracks played,
• generating initial sessions by assigning session identifications to tracks of the determined groups that are played within a specified time difference threshold, and
• determining, from the generated initial sessions, the plurality of sessions that include greater than or equal to a minimum number of tracks played;
• identify, from the listening data, tracks that have been played more than a specified play threshold that is greater than one;
• retain, for the plurality of sessions, the identified tracks;
• retain, from the plurality of sessions, sessions that are greater than a minimum session length threshold and less than a maximum session length threshold;
• generate, from the retained sessions, input-output track sequences, wherein an input to an input-output track sequence includes a sequence of tracks of a specified length and an output of the input-output track sequence includes a next track present in an associated session;
• assign unique identifiers to each of the tracks present across the retained sessions;
• vectorize each input-output track sequence based on associated unique identifiers;
• train a neural network model based on vectorized input-output track sequences, wherein the trained neural network model predicts a next track by generating a unique identifier that corresponds to a track;
• obtain an input playlist;
• utilize the trained neural network model to generate, based on the input playlist, a modified playlist;
• and
• implement a heuristic based path finding model to
• identify, from a user-song interaction graph, tracks that are present in the input playlist,
• identify, from the user-song interaction graph and for the tracks that are present in the input playlist, corresponding edges,
• generate a subgraph from the user-song interaction graph, wherein the subgraph includes the identified tracks that are present in the input playlist and the identified corresponding edges,
• identify, in the subgraph, a track including a highest track popularity based on a composite score,
• identify, from the identified track including the highest track popularity, a next track based on a highest weight, and
• recursively identify a further next track until identification of an end track that does not include a further outgoing edge.