Multi-Dimensional Document Retrieval for Domain-Specific Answers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document question answering systems struggle with low accuracy in retrieving relevant and high-quality documents to provide accurate answers, particularly in specific fields like finance, due to insufficient use of multi-dimensional information such as semantic vectors, event categories, and argument extraction.
Innovation Solution
An answer information generation method using a large language model that involves obtaining semantic vectors and event information from user questions, retrieving candidate documents based on these vectors and categories, determining quality evaluation information, and selecting target documents based on correlation and quality scores to enhance document recall accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional document retrieval methods are used, then the system is simple to operate, but the accuracy of retrieving relevant documents is low
Solution Approach 1:
The patent transforms single-dimensional keyword matching into multi-dimensional retrieval by incorporating semantic vectors, event categories, and argument information as additional retrieval dimensions, thereby improving document retrieval accuracy through enhanced information representation
Solution Approach 2:
The patent segments the question text into multiple components including semantic vectors, event categories, and argument information, allowing each component to contribute independently to the retrieval process, which improves both accuracy and enables systematic handling of complex queries
2Reliability
If multi-dimensional information is used for document retrieval, then the accuracy of answer generation is improved, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary extraction of event categories and argument information from the question text before the retrieval process, preparing structured multi-dimensional information in advance that guides the document selection and improves answer accuracy without adding complexity during the main processing stage
Solution Approach 2:
The patent introduces event categories and argument information as intermediary elements that bridge the question and candidate documents, enabling more precise matching while distributing the computational workload across multiple processing stages rather than concentrating complexity in one operation
3Measurement precision
If quality evaluation is performed on candidate documents, then the relevance of retrieved documents is improved, but the processing time is increased
Solution Approach 1:
The patent applies quality evaluation selectively to specific aspects of candidate documents based on event categories and argument information, focusing assessment on locally relevant qualities rather than performing comprehensive full-document analysis, thereby improving document quality assessment efficiency
Solution Approach 2:
The patent performs partial quality evaluation on candidate documents by focusing on event category matching and argument information presence rather than comprehensive document analysis, achieving sufficient quality assessment for answer generation without the full time cost of complete document evaluation
Data Source
AI summary
A method is provided. The method includes: obtaining, in response to receiving a question text from a user, a semantic vector of the question text and event information related to a specific field; obtaining a plurality of candidate documents from a document library of the specific field based on at least two of the semantic vector of the question text, the at least one piece of argument information and the event category; determining quality evaluation information for each candidate document in the plurality of candidate documents based on the event category; and determining at least one target document from the plurality of candidate documents based on the quality evaluation information of each candidate document and a correlation between each candidate document and the question text, to obtain, based on the at least one target document, answer information used to answer the question text.