Similarity-Based Data Classification With Feedback Reclassification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional classification systems for structured data lack interpretability and scalability, struggle with user feedback integration, and are ill-equipped to handle schema evolution and data drift, leading to reduced trust and inefficiency in data management.
Innovation Solution
A system using similarity-based signatures and multi-level embeddings for clustering and reclassification, with user feedback mechanisms for adaptive self-calibration, enabling precise data grouping and iterative refinement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional classification systems use complex statistical or machine learning models, then classification accuracy is improved, but interpretability deteriorates
Solution Approach 1:
The patent introduces similarity metrics and cluster assignments as intermediary representations between the complex ML model and the user. Instead of presenting raw model predictions, the system computes similarity-based cluster assignments that serve as an interpretable bridge, allowing users to understand classification decisions through similarity relationships without exposing the complex underlying model mechanics.
Solution Approach 2:
The system implements feedback loops where user corrections to cluster assignments are incorporated to iteratively refine the classification model. This feedback mechanism allows the system to learn from interpretability issues and improve both accuracy and interpretability over time by adjusting based on user understanding and corrections.
2Loss of information
If manual labelling or rule-based updates are used for classification, then interpretability is improved, but scalability deteriorates
Solution Approach 1:
The patent segments the classification process into automated similarity-based clustering for scalability and user feedback mechanisms for interpretability. By dividing the task into computational clustering (handling scale) and user review/correction (handling interpretability), the system achieves both scalability and interpretability simultaneously rather than forcing a choice between them.
Solution Approach 2:
The system enables self-service classification through automated similarity-based clustering that operates without manual intervention for routine tasks. Users only need to review and correct exceptional cases, allowing the system to scale automatically while maintaining interpretability through selective user engagement rather than requiring continuous manual oversight.
3Measurement precision
If full model retraining is performed to incorporate user feedback, then classification accuracy is improved, but processing time deteriorates
Solution Approach 1:
Instead of performing complete model retraining, the patent applies partial updates by incorporating user feedback into the existing classification model through iterative refinement. This partial action approach updates only the necessary components based on feedback rather than retraining the entire model, achieving accuracy improvement with significantly reduced processing time.
Solution Approach 2:
The system implements dynamic adaptation where the classification model can be incrementally updated based on user feedback without requiring static full retraining cycles. This dynamic approach allows the model to evolve continuously with small, manageable updates rather than undergoing periodic complete retraining, reducing time loss while maintaining accuracy.
4Productivity
If conventional classification systems are used, then initial classification speed is improved, but adaptability to schema evolution and data drift deteriorates
Solution Approach 1:
The patent implements continuous feedback loops where user corrections and performance monitoring enable the system to detect schema evolution and data drift. This feedback mechanism triggers adaptive responses that update the classification model to accommodate changes, maintaining both speed and adaptability by only retraining when necessary rather than continuously or never.
Solution Approach 2:
The system transitions from static conventional classification to dynamic adaptive classification where the model can evolve with the data. By incorporating user feedback and performance monitoring, the system dynamically adjusts to schema evolution and data drift while maintaining classification speed through incremental updates rather than complete retraining cycles.
Data Source
AI summary
Embodiments of the present disclosure relate to a system and method for classification and reclassification of structured and unstructured data using similarity-based signatures. Entities within a text document of structured and unstructured data are detected by a pre-trained artificial intelligence model. Multi-level embeddings are generated for each entity to capture contextual relationships, enabling calculation of similarity metrics and generation of similarity-based signatures. The entities are clustered based on the embeddings for purposes including visualization and batch classification. Clustering is performed in a first mode based on header information and data types, and in a second mode based on semantic meaning and format characteristics of column data. A user interface enables users to provide feedback on the clustering results, identifying cluster assignments as true positives or false positives. Based on the user feedback, the system reclassifies at least one entity, iteratively refining the AI model and enabling adaptive self-calibration for structured data management.