Document Classification Ensemble Using Multi-Edge Network Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document classification models face limitations due to budget constraints and the need for expert review, leading to inefficiencies and reduced accuracy, particularly in handling diverse and long-tail documents.
Innovation Solution
A computer-implemented method that represents documents as a multi-edge network, combining weak models with a local ensemble model to optimize document classification, allowing for selective manual review and dynamic weight adjustments based on expert input, thereby improving accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a dedicated classification model fine-tuned to company specifics is used, then classification precision is improved, but device complexity and resource requirements increase
Solution Approach 1:
The classification system is segmented into multiple weak models (WMs) that each handle specific aspects of document classification. Instead of using one complex dedicated model, the system divides the classification task across multiple simpler models that can be independently trained and combined, reducing overall system complexity while maintaining precision.
Solution Approach 2:
Multiple weak models are merged into a unified classification system through ensemble methods. The predictions from individual weak models are combined to produce the final classification, achieving high precision through collective decision-making while keeping each individual model simple and manageable.
2Measurement precision
If manual review by expert reviewers is performed on all documents, then classification accuracy is improved, but productivity and time consumption increase
Solution Approach 1:
Instead of performing manual review on all documents, the system applies partial action by selectively reviewing only those documents that fall below a confidence threshold or are identified as high-value targets. This approach achieves sufficient accuracy for critical cases while maintaining high productivity by automating the classification of routine documents.
Solution Approach 2:
The system implements feedback loops where expert reviewers validate and correct classifications, and these corrections are used to retrain and improve the weak models. This continuous feedback mechanism allows the system to progressively improve accuracy while reducing the proportion of documents requiring manual review over time.
3Speed
If weak models are used for initial classification, then processing speed is improved, but measurement precision deteriorates
Solution Approach 1:
Weak models perform preliminary classification of all documents before any manual review or further processing. This preliminary action quickly identifies obvious cases and filters out documents that don't require intensive analysis, enabling the system to process large volumes of documents at high speed while directing attention to only those cases needing higher precision.
Solution Approach 2:
The system adds a temporal dimension to the classification process by using weak models for initial rapid classification, then selectively applying stronger methods only when needed. This multi-stage approach with different precision levels at different stages resolves the contradiction between speed and accuracy by operating in different dimensional regimes.
Data Source
AI summary
Disclosed are techniques that generate a structural representation of a plurality of documents, the structural representation including a plurality of nodes and a plurality of edges, with the plurality of nodes being representations of the plurality of documents and the plurality of edges representing a feature in common between nodes of the plurality of nodes, with each node holding a vector of confidence values for weak models on a current optimization step and a weighted prediction for each of the weak models, generate a local ensemble model from the structural representation of the plurality of documents combined with the weighted prediction of the weak models, with the generated local ensemble model having a higher predictive power than any weak model individually, and generate a label for each node based on the local ensemble model.