Automatic keyphrase labeling using search queries

A system configured to perform automatic keyphrase labeling comprising:

• at least one memory configured to store at least one search query;
• and
• a processor in communication with the at least one memory and configured to:
• extract a plurality of keywords from the at least one search query that resulted in a selection of a document included in a search result;
• for each of the plurality of keywords, determine a probability that the keyword describes the document;
• generate one or more keyphrases by performing processing including selecting each of the plurality of keywords having a probability greater than a predetermined threshold value for insertion into at least one of the one or more keyphrases and assembling the one or more keyphrases from the selected plurality of keywords;
• label the document with the keyphrase;
• and
• store the labeled document as training data in a document corpus in the at least one memory.

A method of automatic keyphrase labeling and machine learning training comprising:

• extracting, by a processor, a plurality of keywords from at least one search query that resulted in a selection of a document of a plurality of documents included in a search result, the extracting comprising removing at least one uninformative portion from the at least one search query, wherein the plurality of keywords comprises a portion of the at least one search query that remains after the at least one uninformative portion is removed;
• determining, by the processor, an affinity between the search query and the document, wherein determining the affinity comprises:
• determining an expected random distribution of selections of the document among selections of any of the plurality of documents;
• determining an actual distribution of selections of the document;
• and
• calculating the affinity based on a proportion of the actual distribution to the expected random distribution;
• for each of the plurality of keywords, determining, by the processor, a probability that the keyword describes the document, the determining comprising:
• determining a keyword component of the probability based on a comparison between the keyword and content of the document;
• and
• calculating the probability using:
• the affinity between the search query and the document;
• and
• the keyword component of the probability;
• generating, by the processor, one or more keyphrases, the generating comprising:
• selecting each of the plurality of keywords having a probability greater than a predetermined threshold value for insertion into at least one of the one or more keyphrases;
• assigning a probability to each word in the at least one uninformative portion;
• and
• assembling the one or more keyphrases from the selected plurality of keywords and each word in the at least one uninformative portion with probability assigned into the keyphrase for the document;
• determining, by the processor, an overall probability that the keyphrase describes the document;
• labeling, by the processor, the document with the keyphrase, the labeling comprising determining that the overall probability is greater than a predetermined overall threshold value;
• and
• training, by the processor, a supervised machine learning model using the labeled document as training data.