Methods, systems, and media for determining playlist title coherence and quality
Methods, systems, and media for determining playlist title coherence and quality are provided. In some embodiments, a method for generating playlist recommendations includes: determining, using a hardware processor, a title of a playlist; generating, using the hardware processor, a byte-level representation of the title based on the title of the playlist; determining, using the hardware processor, an embedded representation of the title based on the byte-level representation; determining, using the hardware processor, a perplexity score of the title by inputting the embedded representation of the title into a trained language model, wherein the perplexity score is an output of the trained language model; and causing, using the hardware processor, a recommendation based on the perplexity score of the title to be presented.
1. A method comprising:
detecting, using a hardware processor, a user-generated title associated with a playlist available on a streaming service;
generating, using the hardware processor, a byte-level representation of the user-generated title based on a sequence of bytes corresponding to the user-generated title of the playlist;
determining, using the hardware processor, an embedded representation of the user-generated title based on the byte-level representation;
determining, using the hardware processor, a plurality of perplexity scores that each correspond to an individual perplexity score of a portion of the embedded representation of the user-generated title by inputting the embedded representation of the user-generated title into a trained language model, wherein the plurality of perplexity scores is an output of the trained language model;
determining, using the hardware processor, an overall perplexity score of the user-generated title based on the plurality of perplexity scores, a number of bytes in the byte-level representation, and a position of each byte in the byte-level representation, wherein the overall perplexity score indicates whether the sequence of bytes corresponding to the user-generated title of the playlist is an unlikely sequence of bytes or a likely sequence of bytes; and
responsive to determining that the overall perplexity score indicates that the sequence of bytes corresponding to the user-generated title of the playlist is a likely sequence of bytes, causing, using the hardware processor, a recommendation system associated with the streaming service to recommend the playlist based on the overall perplexity score of the user-generated title.
2. The method of claim 1 , wherein the trained language model is trained using a plurality of training data that includes at least one of a plurality of pre-approved professionally curated playlist titles and a plurality of pre-approved professionally curated playlist descriptions.
3. The method of claim 1 , wherein the trained language model is trained using a plurality of training data that includes a plurality of pre-approved professionally curated playlist titles, and wherein the trained language model is configured to generate the plurality of perplexity scores of the user-generated title of the playlist based on a negative log-likelihood that the sequence of bytes corresponding to the user-generated title of the playlist would appear in one of the plurality of pre-approved professionally curated playlist titles in the plurality of training data.
4. The method of claim 1 , further comprising:
responsive to determining that the overall perplexity score indicates that the sequence of bytes corresponding to the user-generated title is an unlikely sequence of bytes, causing, using the hardware processor, the recommendation system associated with the streaming service to exclude the playlist from being recommended.
5. The method of claim 1 , further comprising determining that the user-generated title is based in a particular language from a plurality of languages.
6. The method of claim 1 , wherein the playlist comprises a plurality of music tracks.
7. A system for generating playlist recommendations, the system comprising:
a hardware processor that is configured to:
detect a user-generated title associated with a playlist available on a streaming service;
generate a byte-level representation of the user-generated title based on a sequence of bytes corresponding to the user-generated title of the playlist;
determine an embedded representation of the user-generated title based on the byte-level representation;
determine a plurality of perplexity scores that each correspond to an individual perplexity score of a portion of the embedded representation of the user-generated title by inputting the embedded representation of the user-generated title into a trained language model, wherein the plurality of perplexity scores is an output of the trained language model;
determine an overall perplexity score of the user-generated title based on the plurality of perplexity scores, a number of bytes in the byte-level representation, and a position of each byte in the byte-level representation, wherein the overall perplexity score indicates whether the sequence of bytes corresponding to the user-generated title of the playlist is an unlikely sequence of bytes or a likely sequence of bytes; and
responsive to determining that the overall perplexity score indicates that the sequence of bytes corresponding to the user-generated title of the playlist is a likely sequence of bytes, cause a recommendation system associated with the streaming service to recommend the playlist based on the overall perplexity score of the user-generated title.
8. The system of claim 7 , wherein the trained language model is trained using a plurality of training data that includes at least one of a plurality of pre-approved professionally curated playlist titles and a plurality of pre-approved professionally curated playlist descriptions.
9. The system of claim 7 , wherein the trained language model is trained using a plurality of training data that includes a plurality of pre-approved professionally curated playlist titles and wherein the trained language model is configured to generate the plurality of perplexity scores of the user-generated title of the playlist based on a negative log-likelihood that the sequence of bytes corresponding to the user-generated title of the playlist would appear in one of the plurality of pre-approved professionally curated playlist titles in the plurality of training data.
10. The system of claim 9 , wherein the hardware processor is further configured to:
responsive to determining that the overall perplexity score indicates that the sequence of bytes corresponding to the user-generated title is an unlikely sequence of bytes, cause the recommendation system associated with the streaming service to exclude the playlist from being recommended.
11. The system of claim 7 , wherein the hardware processor is further configured to determine that the user-generated title is based in a particular language from a plurality of languages.
12. The system of claim 7 , wherein the playlist comprises a plurality of music tracks.
13. A non-transitory computer-readable medium containing computer executable instructions that, when executed by a processor, cause the processor to:
detect a user-generated title associated with a playlist available on a streaming service;
generate a byte-level representation of the user-generated title based on the user-generated title of the playlist;
determine an embedded representation of the user-generated title based on the byte-level representation;
determine a plurality of perplexity scores that each correspond to an individual perplexity score of a portion of the embedded representation by inputting the embedded representation of the user-generated title into a trained language model, wherein the plurality of perplexity scores is an output of the trained language model;
determine an overall perplexity score of the user-generated title based on the plurality of perplexity scores, a number of bytes in the byte-level representation, and a position of each byte in the byte-level representation, wherein the overall perplexity score indicates whether a sequence of bytes corresponding to the user-generated title of the playlist is an unlikely sequence of bytes or a likely sequence of bytes; and
responsive to determining that the overall perplexity score indicates that the sequence of bytes corresponding to the user-generated title of the playlist is a likely sequence of bytes. cause a recommendation system associated with the streaming service to recommend the playlist based on the overall perplexity score of the user-generated title.
14. The non-transitory computer-readable medium of claim 13 , wherein the trained language model is trained using a plurality of training data that includes at least one of a plurality of pre-approved professionally curated playlist titles and a plurality of pre-approved professionally curated playlist descriptions.
15. The non-transitory computer-readable medium of claim 13 , wherein the trained language model is trained using a plurality of training data that includes a plurality of pre-approved professionally curated playlist titles and wherein the trained language model is configured to generate the plurality of perplexity scores of the user-generated title of the playlist based on a negative log-likelihood that the sequence of bytes corresponding to the user-generated title of the playlist would appear in one of the plurality of pre-approved professionally curated playlist titles in the plurality of training data.
16. The non-transitory computer-readable medium of claim 13 , wherein the execution of the instructions further causes the processor to:
responsive to determining that the overall perplexity score indicates that the sequence of bytes corresponding to the user-generated title is an unlikely sequence of bytes, cause the recommendation system associated with the streaming service to exclude the playlist from being recommended.
17. The non-transitory computer-readable medium of claim 13 , wherein execution of the instructions further causes the processor to determine that the user-generated title is based in a particular language from a plurality of languages.
18. The non-transitory computer-readable medium of claim 13 , wherein the playlist comprises a plurality of music tracks.