Query-Routed Information Provision for Real-Time Language Model Answers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing language models lack the ability to utilize real-time information, limiting their effectiveness in areas where up-to-date data is crucial, such as securities and finance.
Innovation Solution
An information provision device that utilizes a pre-trained language model to classify user queries and employs database inquiry, document-based answers, or direct input based on query type, enabling real-time information provision through structured query language statements and context data retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a pre-trained language model is used to provide information, then the model can generate natural language responses, but the model cannot utilize real-time information after training
Solution Approach 1:
The system segments the information provision process into multiple stages: query classification by a first language model, selection of appropriate processing schemes (database inquiry, document-based answer, or direct language model response), and execution of the selected scheme. This segmentation allows the system to leverage both pre-trained language models and real-time data sources appropriately.
Solution Approach 2:
The patent introduces an intermediary classification system that mediates between user queries and the language model. The first language model classifies query types and determines whether real-time database inquiry or document-based answers are needed, thereby enabling the pre-trained model to indirectly access real-time information through structured intermediaries.
2Loss of information
If the language model is trained on historical data, then the model has comprehensive knowledge, but the model lacks knowledge after the training time
Solution Approach 1:
The system performs preliminary classification of user queries using the first language model before generating responses. This preliminary action identifies whether a query requires real-time database inquiry, document-based answering, or direct language model processing, allowing the system to proactively select the most appropriate information source based on query characteristics.
Solution Approach 2:
The system dynamically adjusts its information provision strategy based on query classification results. For queries requiring real-time information, the system transitions from static pre-trained model responses to dynamic database inquiry or document retrieval, thereby adapting to the temporal requirements of different information needs.
3Adaptability or versatility
If the system uses multiple processing schemes (database inquiry, document-based answer, direct input), then the system can handle diverse query types, but the system complexity increases
Solution Approach 1:
The first language model serves as a universal classifier that handles multiple query types through a single unified interface. Rather than implementing separate specialized systems for database inquiry, document-based answers, and direct language model responses, the patent uses one multi-functional classification model to route all query types through appropriate processing pipelines.
Data Source
AI summary
An information provision device comprising: a processor; and a memory operatively coupled to the processor, wherein the memory stores instructions that, when executed, cause the processor to: receive user query data; classify a type of the user query data by using a first language model; generate, based on the classified type, a target prompt being one of a first prompt requesting conversion of the user query data into structured data and a second prompt requesting identification of information related to the user query data based on context data; and provide response data corresponding to the user query data by using the target prompt and a second language model.