Automated content tagging with latent dirichlet allocation of contextual word embeddings

A non-transitory, computer-readable medium having instructions stored thereon that are executable by a computing device to perform operations comprising:

• inputting contextual N-grams for content into one or more topic-based deep learning models;
• based, at least in part, on one or more confidence values for outputs of the one or more topic-based deep learning models being below a threshold confidence value,
• training a generative statistical model to generate a plurality of content tags from the contextual N-grams;
• and
• aggregating the plurality of content tags into training data for the one or more topic-based deep learning models to generate updated training data;
• and
• retraining the one or more topic-based deep learning models with the updated training data.

An apparatus comprising:

• a processor;
• and
• a computer-readable medium having instructions stored thereon that are executable by the processor to cause the apparatus to,
• preprocess content to generate a plurality of contextual N-grams, wherein a contextual N-gram is a group of consecutive tokens in the content aggregated together based, at least in part, on proximity of word embeddings for words in the contextual N-gram;
• input the plurality of contextual N-grams into a generative statistical model to generate a weighted vector for each of the plurality of contextual N-grams, wherein each weighted vector comprises likelihood values for m topics;
• select as tags for the content those of the plurality of contextual N-grams with greatest of the likelihood values for the m topics;
• and
• supplement a set of one or more topic-based deep learning models with the tags.