Multi-label Classification Engine for Overlapping Taxonomies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems face challenges in performing multi-label classification, especially in large-scale datasets with overlapping classes, where the probability of classes is almost equal, making it difficult to distinguish between them, and require a system capable of handling non-optimal taxonomies and high volumes of data.
Innovation Solution
A three-tier classification framework using ensemble learning, specifically Support Vector Regression and multinomial classifiers, to predict and correct labels in multi-labeled datasets, with dynamic weighted majority for updating expert weights to improve classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning algorithms are used for multi-label classification, then the system can handle basic classification tasks, but it fails to achieve high precision and recall in large-scale datasets with overlapping classes
Solution Approach 1:
The classification problem is divided into multiple independent binary classification tasks, one for each label. Each binary classifier focuses on distinguishing one specific label from all others, which simplifies the overall multi-label classification problem and improves precision for each individual label prediction
Solution Approach 2:
The patent transforms the multi-label classification problem into a multi-dimensional space by creating separate decision boundaries for each label. This dimensional transformation allows the system to handle overlapping classes more effectively by evaluating multiple classification dimensions simultaneously rather than treating it as a single complex classification task
2Productivity
If the system attempts to classify all documents in large-scale datasets, then comprehensive coverage is achieved, but the computational complexity and processing time increase significantly
Solution Approach 1:
The large-scale dataset is processed by dividing the classification task into multiple independent binary classification problems. This segmentation allows for parallel processing of different labels, improving computational efficiency and throughput while maintaining manageable system complexity for each individual classifier
Solution Approach 2:
The system applies binary classification to each label individually, which may be considered an excessive approach compared to traditional multi-class classification. However, this partial action on each label separately ultimately achieves better overall classification performance and scalability for large-scale datasets
3Adaptability or versatility
If the classification model uses a fixed taxonomy structure, then the system is easier to implement, but it cannot adapt to non-optimal taxonomies and overlapping class relationships
Solution Approach 1:
The patent employs dynamic weighted majority voting where the weights of different binary classifiers are adjusted based on their performance. This dynamic adaptation allows the system to handle non-optimal taxonomies and overlapping classes effectively, as the weighting mechanism automatically adapts to the specific characteristics of the data and taxonomy structure
Solution Approach 2:
The binary classification framework acts as an intermediary layer between the input data and the final multi-label classification output. This intermediary approach provides flexibility in handling various taxonomy structures and overlapping class relationships without requiring a fixed rigid classification hierarchy
Data Source
AI summary
In an example, there is disclosed a computing apparatus, including one or more logic elements, including at least one hardware logic element, comprising a classification engine to: receive a clean multi-labeled dataset comprising a plurality of document each assigned to one or more of a plurality of categories; receive an unclean multi-labeled dataset; and produce a recategorized and cleansed dataset from the unclean multi-labeled dataset, comprising predicting a number of labels {circumflex over (l)} for a document j, and comparing {circumflex over (l)} to an existing number of labels l. There is also disclosed a method of providing a classification engine.