Knowledge base generation

A computing system for generating a knowledge base for facilitating interactions between users and an automated assistant associated with the computing system, the computing system comprising:

• one or more processors;
• and
• at least one memory coupled to the one or more processors and storing instructions that, when executed by the one or more processors, cause the computing system to perform operations including:
• receiving a transmission over a communications network from a computing device, the transmission including a plurality of transcripts of user interactions with the automated assistant or agents associated with the computing system;
• transforming ones of the transcripts into one or more question-and-answer (Q & A) pairs associated with a subject of the corresponding user interaction, wherein the transforming includes:
• generating, using a language model (LM), a question based on a user portion of the respective transcript;
• generating, using the LM, an answer based on an agent portion of the respective transcript;
• and
• determining, using the LM, a subject associated with the question and answer based on the respective transcript;
• and
• embedding ones of the Q & A pairs and their associated subjects as vectors in a vector space for retrieval by the automated assistant during subsequent user interactions associated with similar subjects, wherein the embedding includes:
• vectorizing each of the ones of the Q & A pairs and their associated subject into a corresponding vector embedding in real-time.

A method for using a knowledge base to generate an answer to a user's query, the method performed by one or more processors of an online resource and comprising:

• receiving a query from a user over a communications network coupled to the online resource;
• determining, using a first language model (LM), a subject of the user query;
• vectorizing the determined subject into a subject vector embedding;
• identifying a point in vector space corresponding to the subject vector embedding;
• retrieving, from the knowledge base, one or more vector embeddings based on a similarity between the subject vector embedding and the vector embeddings stored in the knowledge base, the similarity determined based in part on the identified point, and each of the retrieved vector embeddings corresponding to a question-and-answer (Q & A) pair used in generating the knowledge base;
• reconstructing the corresponding Q & A pairs from the one or more vector embeddings retrieved from the vector space;
• providing the user query and the reconstructed Q & A pairs to a second LM;
• and
• obtaining, from the second LM, an answer to the user query in real-time with receiving the query.