Tutorial recommendation using discourse-level consistency and ontology-based filtering

A method for training a neural network, comprising:

• receiving training data including a source document and a plurality of candidate documents;
• generating a hidden vector representation for each word of the source document;
• removing at least one word from the source document based on the hidden vector representation to obtain a summary document using a summarization network;
• filtering the plurality of candidate documents based on the source document to obtain a plurality of filtered candidate documents;
• comparing the summary document to each of the filtered candidate documents to obtain a ranking score for each of the filtered candidate documents using a text classification network;
• identifying a relevant candidate document from the filtered candidate documents based on the ranking score;
• computing a loss function using contrastive learning;
• and
• updating parameters of the summarization network based on the loss function.

An apparatus, comprising:

• a word embedding component configured to generate a hidden vector representation for each word of a source document;
• a summarization network configured to remove at least one word from the source document based on the hidden vector representation to obtain a summary document;
• a filtering component configured to filter a plurality of candidate documents based on the source document to obtain a plurality of filtered candidate documents;
• and
• a ranking network configured to compare the summary document to each of the filtered candidate documents to obtain a ranking score for each of the filtered candidate documents.