Semantic Medical Query Responses Through BERT-Based Data Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical experts spend significant time documenting and retrieving patient information from EMR systems, which are not organized effectively, leading to inefficiencies and wasted time due to the lack of meaningful responses to queries.
Innovation Solution
A system and method that classifies unstructured medical data into three types using machine learning models (BERT-based) and stores them in separate databases, enabling retrieval and generation of semantic responses to queries based on these classified data types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If medical information is stored in physical form (papers and folders), then storage is simple, but retrieval time is excessive and information is not easily accessible
Solution Approach 1:
The patent segments medical information into structured categories (patient demographics, clinical findings, prescriptions, lab results) and stores them in separate database tables. This segmentation enables targeted retrieval of specific information types, dramatically reducing retrieval time while maintaining manageable system complexity through modular database design.
Solution Approach 2:
The patent introduces an NLP-based intermediary layer that translates unstructured queries into structured database queries. This intermediary automatically parses user intent, extracts relevant entities, and retrieves information without requiring manual browsing, thus reducing retrieval time while the automated processing keeps complexity manageable.
2Loss of time
If medical experts manually browse through data, then complete information can be found, but significant time is wasted
Solution Approach 1:
The patent implements self-service through automated NLP processing that independently analyzes unstructured queries, identifies relevant information needs, and retrieves data from the database without requiring manual navigation. The system serves itself by automatically translating natural language queries into structured retrieval operations, eliminating the time-consuming manual browsing process while maintaining ease of operation through natural language interfaces.
Solution Approach 2:
The patent replaces the mechanical process of manual data browsing with an automated NLP-based information retrieval system. Instead of physically or manually searching through records, the system uses computational language processing to automatically interpret queries and extract relevant information, significantly reducing retrieval time while maintaining accessibility through natural language interactions.
3Loss of information
If information is displayed without organization, then all data is available, but meaningful responses cannot be generated
Solution Approach 1:
The patent segments medical information into logically organized categories (patient demographics, clinical findings, prescriptions, lab results) with defined relationships between them. This structured segmentation enables the system to retrieve and present meaningful, contextually relevant information in response to queries, rather than displaying unorganized data, while the modular structure keeps management complexity manageable.
Solution Approach 2:
The patent transforms unstructured medical text into structured data with defined parameters and relationships. By changing the organizational parameters from unstructured free text to structured database fields with specific data types and relationships, the system can generate meaningful, contextually appropriate responses while the standardized parameter structure simplifies rather than complicates data management.
4Productivity
If unstructured data is stored, then data entry is simple, but data retrieval and processing are inefficient
Solution Approach 1:
The patent segments medical records into structured components (patient information, clinical findings, prescriptions, laboratory results) stored in separate database tables with defined relationships. This segmentation dramatically improves data processing efficiency by enabling targeted queries and automated analysis, while the modular structured design keeps complexity manageable through standardized schemas and relationships.
Solution Approach 2:
The patent replaces simple unstructured text storage with a structured database system that uses automated NLP processing for data entry and retrieval. The structured format enables efficient computational processing and automated information extraction, significantly improving productivity while the systematic structure provides clear organization that simplifies rather than complicates data management.
Data Source
AI summary
The implementations of the present disclosure provide a method and a system for generating a semantic response to a query. The method comprises receiving unstructured continuous data of a user, classifying the unstructured continuous data into a first type of data, by a first classification model, a second type of data, by a second classification model, and a third type of data, by a third classification model, wherein the second and third type of data includes name related data and event related data respectively. The method further comprises storing the first, second and third type of data in a first, a second and a third database respectively, and receiving a query to search for one or more details corresponding to the user, wherein the one or more details are based on the unstructured continuous data of the user.