Inter-document search using knowledge-enriched vectors

An apparatus, comprising:

• at least one processor;
• and
• at least one memory storing instructions that, when executed by the at least one processor, cause the at least one processor to:
• encode, using an encoding machine learning model, in real time, a contextualized embedding for a search query to form a search vector, the search query includes at least one search term and seeks information within a plurality of electronic documents, the contextualized embedding includes a vector representation of the at least one search term having contextual information for the at least one search term;
• in response to receiving the search query, generate a knowledge graph structure using the at least one search term, wherein the at least one search term is associated with at least another term in a plurality of terms in the knowledge graph structure, wherein the knowledge graph structure includes a base knowledge graph structure generated using another machine learning model based on the plurality of electronic documents prior to the receiving and the at least one search term;
• modify, using the encoding machine learning model, in real time, the contextualized embedding for the search query using the at least another term to generate a modified contextualized embedding, wherein the modified contextualized embedding includes encoded one or more terms from the knowledge graph structure and encoded one or more connections between the one or more terms and with the at least one search term from the knowledge graph structure and the at least one search term, and generate a modified search vector using the search vector and the modified contextualized embedding;
• and
• search, in real-time, the plurality of electronic documents using the modified search vector and generate a response to the search query based on one or more responsive document vectors retrieved from the plurality of electronic documents and that are semantically similar to the modified search vector.

A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by at least one processor, cause the at least one processor to:

• generate, in response to receiving of one or more queries, a knowledge graph structure using at least one search term, wherein the at least one search term is associated with at least another term in a plurality of terms in the knowledge graph structure, wherein the at least one search term is associated with the one or more queries seeking information in a plurality of electronic documents, wherein the knowledge graph structure includes a base knowledge graph structure generated using another machine learning model based on the plurality of electronic documents prior to the receiving and the at least one search term;
• form a search vector, using a contextualized embedding generated, using the one or more queries and the at least another term, wherein formation of the search vector includes encoding, using an encoding machine learning model, in real time, the contextualized embedding to form the search vector, the contextualized embedding includes a vector representation of the at least one search term having contextual information for the at least one search term;
• wherein the contextualized embedding is modified, using the encoding machine learning model, in real time, using the at least another term to generate a modified contextualized embedding, wherein the modified contextualized embedding includes encoded one or more terms from the knowledge graph structure and encoded one or more connections between the one or more terms and with the at least one search term from the knowledge graph structure and the at least one search term, and a modified search vector is generated using the search vector and the modified contextualized embedding;
• and
• search, in real-time, the plurality of electronic documents using the modified search vector and generate a response to the one or more queries based on one or more responsive document vectors retrieved from the plurality of electronic documents and that are semantically similar to the modified search vector.