Natural Language Domain Determination Through Vector Intent Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large language models often misinterpret user queries in enterprise environments, leading to irrelevant or incorrect responses due to the inability to accurately understand the user's intent.
Innovation Solution
A system and method that utilizes a natural language understanding (NLU) model to generate a domain intent list with vector indices, selects relevant vector structures from a vector store, and uses an answer generation model to provide accurate responses by predicting user intent and utilizing embeddings with high similarity scores, with continuous feedback training to improve performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a large language model is used to interpret user requests, then the system can handle natural language queries, but the model may mischaracterize the user's intent leading to irrelevant or wrong responses
Solution Approach 1:
The patent introduces an intermediary NLU engine between the user query and the answer generation model. This NLU engine processes the user query through embedding models and vector structures to determine domain intent, acting as a mediator that improves intent understanding accuracy before the query reaches the answer generation model.
Solution Approach 2:
The system segments the answer generation process into distinct components: an embedding model that converts queries to vectors, an NLU engine that processes vectors to determine intent, and an answer generation model that produces responses. This segmentation allows each component to specialize in specific tasks, improving overall accuracy.
2Measurement precision
If the system processes user queries through multiple components (embedding model, NLU engine, vector store), then the accuracy of intent determination is improved, but the system complexity increases
Solution Approach 1:
The NLU engine serves multiple functions within the system: it processes user queries, generates domain intent lists, and interacts with the vector store. This multi-functionality reduces the need for separate specialized components, thereby managing system complexity while maintaining high accuracy.
Solution Approach 2:
The system performs preliminary processing of user queries through the embedding model and NLU engine before the main answer generation occurs. By pre-processing queries to determine domain intent in advance, the system reduces the complexity of the main answer generation task.
3Measurement precision
If the system uses vector embeddings and similarity scoring to determine relevant information, then the relevance of responses is improved, but the processing time and computational resources increase
Solution Approach 1:
The system performs partial processing by generating only the necessary portion of the query embedding and selecting only the most relevant vector structures from the vector store. This partial action approach maintains high relevance accuracy while reducing unnecessary computational overhead and processing time.
Data Source
AI summary
A method includes generating a user query embedding for a user query received from a user, generating a domain intent list comprising at least one vector index, selecting at least one vector structure corresponding to the at least one vector index to obtain a set of selected vector structures from a plurality of vector structures in a vector store, obtaining at least one result embedding wherein the at least one result embedding matches the user query embedding, transmitting the user query and the at least one result embedding to an answer generation model and receiving the answer to the user query.