Machine learning for associating skills with content

A method comprising:

• generating a first skill embedding based on first text of a first skill tag of a skill taxonomy that is based on one or more user-provided skills;
• extracting a first plurality of tokens from a first content item that includes at least one of audio and video and does not have an associated transcript;
• generating a first plurality of token embeddings based on the first plurality of tokens;
• for each token embedding of the first plurality of token embeddings:
• computing a first plurality of measurements between the first skill embedding and said each token embedding;
• based on the first plurality of measurements, applying a first set of values to a classification model trained using training data that correspond to the first skill tag;
• generating, by the classification model, a confidence score;
• and
• when the confidence score is greater than a particular threshold, storing first skill tagging data that associate the first skill tag with the first content item;
• wherein the method is performed by one or more computing devices.

One or more storage media storing instructions which, when executed by one or more processors, cause:

• training a plurality of classification models, wherein the plurality of classification models comprises a separate classification model for each skill tag of a plurality of different skill tags, an output of each classification model of the plurality of classification models corresponds to a different skill tag of the plurality of skill tags, and each classification model is trained based on textual embeddings of a different plurality of content items and labels indicating whether each content item is associated with the different skill tag that corresponds to said each classification model;
• when a length of text associated with a particular content item that contains at least one of audio and video is greater than a threshold, (i) generating a particular content item embedding based on the text from the particular content item, (ii) applying the particular content item embedding to each of the plurality of classification models to generate a plurality of results, (iii) identifying one or more results of the plurality of results that indicate that one or more corresponding skill tags are associated with the particular content item, and (iv) for each result of the one or more results, storing skill tagging data that associates the particular content item with a particular skill tag that corresponds to said each result.