Cold-start podcast recommendations

1. A method for providing podcast recommendations, the method comprising:

receiving a request for a podcast recommendation;

determining whether a profile of the user indicates consumption by the user of podcasts;

responsive to determining that the profile does not indicate consumption by the user of podcasts:

determining music preferences of the user based on data of the profile;

generating a podcast seed based on the music preferences by:

determining a prominent attribute of the music preferences, the prominent attribute being an artist attribute;

accessing a knowledge graph that maps attributes of the music preferences of users to podcast references;

identifying, from the knowledge graph, a podcast preference that is mapped to the prominent attribute, the podcast preference reflecting a podcast that is played back by above a threshold percentage of users that prefer the artist; and

updating the podcast seed to include the podcast preference as an attribute of the podcast seed;

selecting a podcast recommendation for the user based on the podcast seed, the selected podcast recommendation being for the podcast reflected by the podcast preference; and

providing the podcast recommendation to the user;

receiving a selection of the podcast recommendation by the user; and

responsive to receiving the selection, playing back a podcast corresponding to the podcast recommendation.

2. The method of claim 1 , wherein generating the podcast seed based on the music preferences comprises:

determining a category of music that is preferred by the user based on the music preferences; and

updating the podcast seed to include the category as an attribute of the seed.

3. The method of claim 1 , wherein selecting the podcast recommendation for the user based on the podcast seed comprises:

identifying a series of podcasts;

determining an episode of the series that, relative to other episodes of the series, most closely matches the podcast seed; and

selecting the episode as the podcast recommendation.

4. The method of claim 3 , wherein the episode is later in the series than a first episode of the series.

5. The method of claim 4 , wherein selecting the episode of the podcast recommendation comprises:

determining whether the episodes are labeled for sequential playback;

responsive to determining that the episodes are not labeled for sequential playback, selecting the episode as the podcast recommendation; and

responsive to determining that the episodes are labeled for sequential playback, selecting the first episode as the podcast recommendation.

6. The method of claim 5 , wherein the first episode is selected as the podcast recommendation, and wherein providing the podcast recommendation to the user comprises including, with the podcast recommendation, an indication of the episode as a reason for which the podcast recommendation is being made.

7. A method for providing podcast recommendations, the method comprising:

receiving a request for a podcast recommendation;

determining whether a profile of the user indicates consumption by the user of podcasts;

responsive to determining that the profile does not indicate consumption by the user of podcasts:

determining music preferences of the user based on data of the profile;

generating a podcast seed based on the music preferences;

selecting a podcast recommendation for the user based on the podcast by:

identifying a podcast that has attributes corresponding to the podcast seed;

identifying a plurality of similar podcasts to the identified podcast by:

identifying a social media page corresponding to the identified podcast;

identifying a plurality of users who subscribe to the social media page;

searching a provider of the social media page for another social media page corresponding to another podcast to which a threshold percentage of the plurality of users also subscribe; and

responsive to locating, from the searching, the another social media page, adding the another podcast to the plurality of similar podcasts; and

generating a playlist including the plurality of similar podcasts; and

providing the podcast recommendation to the user;

receiving a selection of the podcast recommendation by the user; and

responsive to receiving the selection, playing back a podcast corresponding to the podcast recommendation.

8. A non-transitory computer readable medium configured to store instructions, the instructions when executed by a processor cause the processor to perform operations, the instructions comprising instructions to:

receive a request for a podcast recommendation;

determine whether a profile of the user indicates consumption by the user of podcasts;

responsive to determining that the profile does not indicate consumption by the user of podcasts:

determine music preferences of the user based on data of the profile;

generate a podcast seed based on the music preferences;

select a podcast recommendation for the user based on the podcast seed by:

identifying a series of podcasts;

determining an episode of the series that, relative to other episodes of the series, most closely matches the podcast seed, wherein the episode is later in the series than a first episode of the series; and

selecting the episode as the podcast recommendation by:

determining whether the episodes are labeled for sequential playback;

responsive to determining that the episodes are not labeled for sequential playback, selecting the episode as the podcast recommendation; and

responsive to determining that the episodes are labeled for sequential playback, selecting the first episode as the podcast recommendation; and

provide the podcast recommendation to the user;

receive a selection of the podcast recommendation by the user; and

responsive to receiving the selection, playing back a podcast corresponding to the podcast recommendation.

9. The non-transitory computer readable medium of claim 8 , wherein the instructions to generate the podcast seed based on the music preferences comprise instructions to:

determine a category of music that is preferred by the user based on the music preferences; and

update the podcast seed to include the category as an attribute of the seed.

10. The non-transitory computer readable medium of claim 8 , wherein the instructions to generate the podcast seed based on the music preferences comprise instructions to:

determine a prominent attribute of the music preferences;

access a knowledge graph that maps attributes of the music preferences of users to podcast preferences;

identify, from the knowledge graph, a podcast preference that is mapped to the prominent attribute; and

update the podcast seed to include the podcast preference as an attribute of the podcast seed.

11. The non-transitory computer readable medium of claim 10 , wherein the prominent attribute is an artist, wherein the podcast preference reflects a podcast that is played back by above a threshold percentage of users that prefer the artist, and wherein the instructions to select the podcast recommendation based on the podcast seed comprise instructions to:

select the podcast to be the podcast that is played back by above the threshold percentage of users that prefer the artist.

12. The non-transitory computer readable medium of claim 8 , wherein the instructions to select the podcast recommendation comprise instructions to:

identify a podcast that has attributes corresponding to the podcast seed;

identify a plurality of similar podcasts to the identified podcast; and

generate a playlist including the plurality of similar podcasts.

13. The non-transitory computer readable medium of claim 12 , wherein the instructions to identify the plurality of similar podcasts to identified podcast comprise instructions to:

identify a social media page corresponding to the identified podcast;

identify a plurality of users who subscribe to the social media page;

search a provider of the social media page for another social media page corresponding to another podcast to which a threshold percentage of the plurality of users also subscribe; and

responsive to locating, from the searching, the another social media page, add the another podcast to the plurality of similar podcasts.

14. The non-transitory computer readable medium of claim 8 , wherein the first episode is selected as the podcast recommendation, and wherein the instructions to provide the podcast recommendation to the user comprise instructions to include, with the podcast recommendation, an indication of the episode as a reason for which the podcast recommendation is being made.