Cross-context natural language model generation

A tangible, non-transitory, machine-readable medium storing instructions that, when executed by one or more processors, effectuate operations comprising:

• obtaining, with a computer system, a set of ontology graphs comprising a first ontology graph associated with a first domain category value and a second ontology graph associated with a second domain category value, wherein vertices of the first ontology graph are connected to vertices of the second ontology graph via a set of graph edges;
• obtaining, with the computer system, a natural-language text document in response to a query made by a user associated with a set of context parameters;
• obtaining, with the computer system, a set of domain category values identified using the set of context parameters, wherein the set of domain category values comprises the second domain category value;
• determining, with the computer system, a first set of embedding vectors in an embedding space based on a first set of n-grams of the natural-language text document, the determining the first set of embedding vectors comprising using a first set of neural network layers to generate the first set of embedding vectors;
• determining a set of hidden state values of the first set of neural network layers, the set of hidden state values representing intermediate outputs that are computed by a first layer of the first set of neural network layers and provided to a second layer of the first set of neural network layers as inputs;
• update the set of hidden state values by performing a set of SoftMax operations, wherein performing the set of SoftMax operations comprises determining a ratio of exponential values computed from the set of hidden state values provided by the second layer;
• obtaining an attention vector;
• determining a weighted sum of the set of hidden state values by multiplying elements of the set of hidden state values by the attention vector;
• retrieving, with the computer system, a first set of vertices of the first ontology graph based on the first set of embedding vectors, wherein each respective vertex of the first set of vertices is identified by a respective embedding vector of the first set of vertices;
• selecting, with the computer system, a second set of vertices of the second ontology graph, wherein the second set of vertices is adjacent to the first set of vertices via the set of graph edges associating the first ontology graph and the second ontology graph;
• generating, with the computer system using the updated set of hidden state values, a sequence of n-grams based on a second set of embedding vectors identified by the second set of vertices;
• and
• presenting, with the computer system, a user interface (UI) comprising a summary to a visual display.

A system comprising:

• one or more memory devices storing instructions;
• and
• one or more processors configured to execute the instructions that, when executed, cause operations comprising:
• obtaining, with a computer system, a set of ontology graphs comprising a first ontology graph associated with a first domain category value and a second ontology graph associated with a second domain category value, wherein vertices of the first ontology graph are connected to vertices of the second ontology graph via a set of graph edges;
• obtaining, with the computer system, a natural-language text document in response to a query made by a user associated with a set of context parameters;
• obtaining, with the computer system, a set of domain category values identified using the set of context parameters, wherein the set of domain category values comprises the second domain category value;
• determining a first set of embedding vectors in an embedding space based on a first set of n-grams of the natural-language text document, the determining the first set of embedding vectors comprising using a first set of neural network layers to generate the first set of embedding vectors;
• determining a set of hidden state values of the first set of neural network layers, the set of hidden state values representing intermediate outputs that are computed by a first layer of the first set of neural network layers and provided to a second layer of the first set of neural network layers as inputs;
• update the set of hidden state values by performing a set of SoftMax operations, wherein performing the set of SoftMax operations comprises determining a ratio of exponential values computed from the set of hidden state values provided by the second layer;
• obtaining an attention vector;
• determining a weighted sum of the set of hidden state values by multiplying elements of the set of hidden state values by the attention vector;
• retrieving a set of associations based on the first domain category value and the second domain category value;
• obtaining a second set of embedding vectors using the set of associations based on the first set of embedding vectors, wherein each respective first embedding vector of the first set of embedding vectors are mapped to a respective second embedding vector of the second set of embedding vectors;
• generating, using the updated set of hidden state values, a sequence of n-grams of a summary based on the second set of embedding vectors using an abstractive text summarization model;
• and
• presenting a user interface comprising the sequence of n-grams to a visual display.