Learning graph representations using hierarchical transformers for content recommendation

A computer-implemented method of predicting a target entity in a knowledge graph comprising:

• receiving, by an entity transformer, a source entity-relation pair input and a neighborhood entity-relation pair input from the knowledge graph;
• capturing, by the entity transformer, interaction information for the source entity-relation pair input and interaction information for the neighborhood entity-relation pair input;
• aggregating, by a context transformer, the interaction information for the source entity-relation pair input and the interaction information for the neighborhood entity-relation pair input;
• outputting, by the context transformer, target entity predictions based on the aggregated interaction information;
• ranking the target entity predictions based on a plausibility score;
• selecting the highest ranked target entity prediction to be the target entity;
• and
• adding the target entity to the knowledge graph.

A hierarchical transformer model for predicting a target entity in a knowledge graph comprising:

• a processor;
• memory storing computer-executable instructions, which when executed, cause the hierarchical transformer model to:
• receive, by a first level transformer block, a source entity-relation pair input and a neighborhood entity-relation pair input from the knowledge graph;
• capture, by the first level transformer block, interaction information for the source entity-relation pair input and interaction information for the neighborhood entity-relation pair input;
• aggregate, by a second level transformer block, the interaction information for the source entity-relation pair input and the interaction information for the neighborhood entity-relation pair input;
• output, by the second level transformer block, link predictions based on the aggregated interaction information;
• rank the link predictions based on a plausibility score;
• select the highest ranked link prediction to be the target entity for the knowledge graph;
• and
• add the target entity to the knowledge graph.