Hyper-graph learner for natural language comprehension

A server computer comprising:

• a processor;
• a memory device;
• and
• a computer-readable medium coupled to the processor, the computer-readable medium comprising code executable by the processor for implementing a method comprising:
• training a chat bot, wherein the training includes:
• retrieving text data from a conversation database;
• parsing the text data into a first set of distinct words;
• performing using the first set of distinct words, a concurrence calculation by tallying a number of times that different pairs of words from the first set of distinct words occur together in specific sentences from the text data;
• generating a weighted graph, the weighted graph including a plurality of nodes and a plurality of edges between the plurality of nodes, each node in the plurality of nodes representing a word from the first set of distinct words, and each edge in the plurality of edges representing a calculated concurrence between a pair of words from the first set of distinct words;
• performing graph learning using a machine learning algorithm and the weighted graph to generate a plurality of learned communities of language, where each of the plurality of learned communities of language includes a subset group of words from the first set of distinct words,
• determining a plurality of representative vectors of values based on the plurality of learned communities of language, each of the plurality of representative vectors of values corresponding to a learned community of language from the plurality of learned communities of language;
• receiving, using the chat bot, a text phrase;
• predicting, using the chat bot, a response, wherein the predicting includes:
• parsing the text phrase to generate a second set of distinct words;
• generating a second vector of values relating to the second set of distinct words;
• determining, based on the second vector of values, a most mathematically similar vector of values from the plurality of representative vectors of values, where the most mathematically similar vector of values corresponds to a most similar community from the plurality of learned communities of language;
• querying an action set associated with the most similar community;
• and
• generating the response based on the action set;
• and
• providing, using the chat bot, the response.