Native On-Device Field Extraction for Low-Latency Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing field extraction technologies in computer applications are resource-intensive, causing high latency, network bandwidth consumption, and require manual user input, leading to inefficiencies and inaccuracies in processing natural language documents.
Innovation Solution
Implementing native automatic field extraction at a user device using compressed machine learning models and intelligent user interfaces that score and rank values relative to keywords, eliminating the need for remote computing and manual input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If remote natural language processing services are used to determine semantic meaning of words, then processing capability is improved, but latency and network bandwidth consumption increase
Solution Approach 1:
The patent extracts the natural language processing capability from remote servers and embeds it directly into the mobile device through a trained model. This allows the device to perform semantic analysis locally without needing to communicate with remote services, thereby eliminating network latency while maintaining processing capability.
Solution Approach 2:
The patent introduces a trained natural language processing model as an intermediary between the document image and the field extraction process. This model acts as a local mediator that performs semantic understanding directly on the device, replacing the need for remote intermediary services and reducing dependency on network infrastructure.
2Reliability
If remote natural language processing services are used, then processing capability is improved, but network bandwidth consumption increases
Solution Approach 1:
The patent extracts the natural language processing capability from remote servers and embeds it directly into the mobile device through a trained model. This allows the device to perform semantic analysis locally without needing to communicate with remote services, thereby eliminating network bandwidth consumption while maintaining processing capability.
Solution Approach 2:
The mobile device becomes self-sufficient by hosting the trained NLP model locally. It performs semantic understanding and field extraction independently without requiring external network services, making the system self-service oriented and eliminating continuous network bandwidth consumption.
3Device complexity
If manual user input is required for field extraction, then system complexity is reduced, but productivity and accuracy decrease
Solution Approach 1:
The patent performs preliminary actions by training the NLP model in advance with labeled data containing field names and their corresponding values. This pre-trained knowledge enables the model to automatically identify and extract fields without requiring manual user input during the actual extraction process, thereby improving productivity while maintaining manageable system complexity.
Solution Approach 2:
The system performs self-service by automatically identifying and extracting fields using the trained model without requiring manual user intervention. The model independently analyzes the document image, determines semantic meanings, and extracts relevant fields, significantly improving productivity while keeping the interface simple.
4Device complexity
If manual user input is required for field extraction, then system complexity is reduced, but accuracy decreases
Solution Approach 1:
The patent performs preliminary training of the NLP model with accurately labeled data that includes field names and their corresponding values. This pre-training establishes accurate semantic understanding and field identification capabilities, enabling the system to achieve high extraction accuracy automatically without requiring manual correction or input during operation.
Solution Approach 2:
The system incorporates feedback mechanisms where the trained model continuously improves its field extraction accuracy based on labeled examples and performance metrics. The model learns from correct and incorrect extractions, refining its semantic understanding and field identification capabilities over time, thereby achieving high accuracy while maintaining simple operation.
Data Source
AI summary
Particular embodiments receive a plurality of values associated with a document. One or more keywords associated with the document are also received. A first score is generated for each value, of the plurality of values. The generation of the first score excluding sending, over a computer network, a first request to one or more remote computing devices to generate the first score. Based at least in part on the generating of the first score, each value, of the plurality of values is ranked. The ranking of each value excluding sending, over the computer network, a second request to the one or more remote computing devices to rank each value. Based on the ranking, at least one value, of the plurality of values, is selected. The selecting being indicative that the at least one value is a candidate to be a constituent of the one or more keywords. Based on the ranking, an indicator indicating that the at least one value is the candidate to be the constituent of the one or more keywords is presented.