Determining repair instructions in response to natural language queries

A computer-implemented method comprising:

• encoding, by one or more computing systems, information from multiple documents about how to perform a plurality of types of at least one of repair actions or maintenance actions in a target area, including:
• separating, by the one or more computing systems and for each of the multiple documents, content from that document into multiple content groups that are each a subset of that content, and associating, for each of the multiple content groups, additional information with that content group to provide expanded content for that content group;
• and
• generating, by the one or more computing systems, a plurality of content embedding vectors to represent the information from the multiple documents, including generating, for each of the multiple content groups of each of the multiple documents, one of the plurality of content embedding vectors to represent semantic information of that content group;
• encoding, by the one or more computing systems, a received query about an indicated type of the at least one of repair actions or maintenance actions in the target area, wherein the received query is provided in a natural language format, and wherein the encoding of the received query includes generating a query embedding vector that represents semantic information of the received query;
• determining, by the one or more computing systems, a response to the received query that provides instructions for the indicated type of the at least one of repair actions or maintenance actions, including:
• matching, by the one or more computing systems, the query embedding vector to a group of multiple candidate content embedding vectors, wherein the multiple candidate content embedding vectors are a subset of the plurality of content embedding vectors and each has a matching distance to the query embedding vector below a defined threshold;
• validating, by the one or more computing systems, and for each of one or more identified candidate content embedding vectors from the group of multiple candidate content embedding vectors, that a combination of content from information of the content group represented by that identified candidate content embedding vector and the expanded content for that content group provides the instructions for the indicated type of the at least one of repair actions or maintenance actions;
• and
• generating, by the one or more computing systems, the instructions for the indicated type of the at least one of repair actions or maintenance actions from the combination of content for at least one of the one or more identified candidate content embedding vectors;
• and
• providing, by the one or more computing systems, the determined response to the received query, to initiate the indicated type of the at least one of the repair actions or maintenance actions.

A non-transitory computer-readable medium having stored contents that cause one or more computing systems to perform automated operations, the automated operations including at least:

• obtaining, by the one or more computing systems, encoded information about how to perform a plurality of types of actions in a target area, wherein the encoded information includes a plurality of content embedding vectors with, for each of multiple content groups in the target area, one of the plurality of content embedding vectors to represent semantic information of that content group;
• encoding, by the one or more computing systems, a received query that is about an indicated type of action in the target area and is provided in a natural language format, including generating a query embedding vector that represents semantic information of the received query;
• determining, by the one or more computing systems, a response to the received query that provides information about the indicated type of action, including validating, for each of one or more candidate content embedding vectors that are identified from the plurality of content embedding vectors based at least in part on each matching the query embedding vector, that the sematic information of the content group represented by that candidate content embedding vector provides at least some of the information about the indicated type of action, and obtaining the information about the indicated type of action by, for at least one of the one or more candidate content embedding vectors, using information of at least one content group represented by the at least one candidate content embedding vector;
• and
• providing, by the one or more computing systems, the determined response to the received query including the information about the indicated type of action, to initiate implementation of the indicated type of action.

A system, comprising:

• one or more hardware processors of one or more computing systems;
• and
• one or more memories with stored instructions that, when executed by at least one of the one or more hardware processors, cause the one or more computing systems to perform automated operations including at least:
• obtaining encoded information about how to perform a plurality of types of actions involving at least one of treating or preventing problems in a target area, wherein the encoded information includes a plurality of content embedding vectors with, for each of multiple content groups related to the target area, one of the plurality of content embedding vectors to represent semantic information of that content group;
• encoding a received query that is about an indicated problem in the target area and is provided in a natural language format, and wherein the encoding of the received query includes generating a query embedding vector that represents semantic information of the received query;
• determining a response to the received query that provides information about at least one of treating or preventing the indicated problem, including validating, for each of one or more candidate content embedding vectors that are identified from the plurality of content embedding vectors and that each matches the query embedding vector, that information of the content group represented by that candidate content embedding vector provides at least some of the information about the at least one of the treating or the preventing of the indicated problem, and obtaining the information about the at least one of the treating or the preventing of the indicated problem by, for at least one of the one or more candidate content embedding vectors, using the information in at least one content group represented by the at least one candidate content embedding vector;
• and
• providing the determined response to the received query including the information about the at least one of treating or preventing the indicated problem, to initiate the at least one of the treating or the preventing of the indicated problem.