Identification of synthetic examples for improving search rankings

A system for identifying synthetic media file examples to train a supervised machine learned ranking algorithm to rank relevance of media files to a search query, the system comprising:

• a memory comprising:
• a supervised machine learned ranking algorithm, and a collection of media files;
• and
• a processor configured to execute instructions to:
• identify a search query for a search engine for the collection of media files from a plurality of previously received search queries for the search engine;
• determine a similarity threshold value based on a distance calculation between media files;
• identify a cluster of similar media files within the similarity threshold value;
• select a negative cluster from the cluster of similar media files having a frequency of downloads associated with the search query for the media files that is below a download threshold value;
• select at least one training media file, at random, from the negative cluster as a synthetic negative example related to the search query based at least on at least one training media file having a frequency of downloads below the download threshold value;
• select at least another training media file from the collection of media files as a synthetic positive example for the search query;
• provide, to the supervised machine learned ranking algorithm in order to generate a logistic regression model that provides a probability that a media file is responsive to the search query, a training set comprising:
• an identification of the search query,
• a copy of the at least one training media file,
• a first indicator that the at least one training media file is a synthetic negative example for the search query,
• a copy of the at least another training media file selected as the synthetic positive example for the search query, and
• a second indicator that the at least another training media file is a synthetic positive example for the search query;
• provide, to the supervised machine learned ranking algorithm, the search query and the collection of media files;
• and
• receive, from the supervised machine learned ranking algorithm, a ranking of the collection of media files according to a probability that each media file is responsive to the search query.

A non-transitory machine-readable storage medium comprising machine-readable instructions for causing a processor to execute a method for identifying synthetic media file examples to train a supervised machine learned ranking algorithm to rank relevance of media files to a search query, comprising:

• identifying a search query for a search engine for a collection of media files from a plurality of previously received search queries for the search engine;
• determining a similarity threshold value based on a distance calculation between media files;
• identifying a cluster of similar media files within the similarity threshold value;
• selecting a negative cluster from the cluster of similar media files having a frequency of downloads associated with the search query for the media files that is below a download threshold value;
• selecting at least one training media file, at random, from the negative cluster as a synthetic negative example related to the search query based at least on at least one training media file having a frequency of downloads below the download threshold value;
• selecting at least another training media file from the collection of media files as a synthetic positive example for the search query;
• providing, to a supervised machine learned ranking algorithm in order to generate a logistic regression model that provides a probability that a media file is responsive to the search query, a training set comprising:
• an identification of the search query,
• a copy of the at least one training media file,
• a first indicator that the at least one training media file is a synthetic negative example for the search query,
• a copy of the at least another training media file selected as the synthetic positive example for the search query, and
• a second indicator that the at least another training media file is a synthetic positive example for the search query;
• providing, to the supervised machine learned ranking algorithm, the search query and the collection of media files;
• and
• receiving, from the supervised machine learned ranking algorithm, a ranking of the collection of media files according to a probability that each media file is responsive to the search query.