Classifier Training via Case-Centric Feature Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Processing large datasets from online user activities to extract useful information is challenging due to varying representation techniques and lack of consistent ontologies across different online entities, making it difficult to identify significant information and trends.
Innovation Solution
A system that develops a classifier by grouping raw data into cases, using statistical analysis of features from entire cases rather than individual instances, and automatically selects the most beneficial cases for trainer input to generate a robust classifier capable of identifying semantically-similar data fields across different online entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If statistical analysis is performed on individual data instances, then processing speed is faster, but classification accuracy deteriorates due to lack of contextual information
Solution Approach 1:
The patent combines multiple data instances into a single aggregated data structure that preserves statistical properties. By merging instances and analyzing them collectively rather than individually, the system maintains processing efficiency while improving classification accuracy through enhanced contextual information.
2Reliability
If all available data is used for training, then classifier robustness is improved, but training time and computational resources increase
Solution Approach 1:
The patent extracts and utilizes only the essential statistical properties from data instances rather than processing complete raw data. By extracting key features and statistical characteristics, the system achieves robust classifier training with reduced computational overhead and faster training times.
Solution Approach 2:
The patent applies partial action by using aggregated statistical properties rather than complete instance data. This partial representation captures sufficient information for robust classification while significantly reducing the data volume requiring processing, thereby decreasing training time and resource consumption.
3Measurement precision
If manual labeling of all data is performed, then data quality is improved, but labor costs and processing time increase significantly
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate high-quality training data through statistical aggregation and analysis. The automated process maintains data quality without requiring manual intervention, thereby preserving processing throughput and eliminating labor bottlenecks.
4Adaptability or versatility
If data from multiple online entities is processed, then coverage and versatility are improved, but complexity increases due to varying representation techniques
Solution Approach 1:
The patent creates a universal data representation framework that can handle diverse data sources with varying formats. By establishing a common statistical aggregation approach that works across different online entities and representation techniques, the system achieves broad data coverage without proportionally increasing complexity.
Data Source
AI summary
An exemplary embodiment of the present invention provides a computer implemented method of developing a classifier. The method includes obtaining a set of training data comprising labeled cases. The method also includes training a classifier based, at least in part, on the training data. The method also includes applying the classifier to a plurality of unlabeled cases to generate classification scores for each of the unlabeled cases, wherein each classification score corresponds with an instance of a corresponding case. Furthermore, the classification score corresponding to a first instance in a case is computed based, at least in part, on a value of a case-centric feature corresponding to the first instance, wherein the value of the case-centric feature is based, at least in part, on characteristics of the first instance and a second instance in the case.