Multi-Label SVM Classification Using Hyperplane Distance Tiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-label classification methods, such as support vector machines running a one-vs-the-rest classifier, struggle to effectively assign multiple labels to websites containing content related to multiple topics, especially when the number of categories exceeds 50, leading to a significant proportion of unlabeled documents.
Innovation Solution
A modified multi-label classification system that calculates distances between feature vectors and hyperplanes for each label, allowing for tiered labeling, ensuring that at least one label is assigned by considering positive and negative distances from the mean, using Gaussian distributions and standard deviations to determine relevant labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional one-vs-the-rest classifier is used with large number of label classes, then classification coverage is reduced, but labeling completeness deteriorates
Solution Approach 1:
The patent segments the classification process into multiple tiers. The first tier uses a traditional one-vs-the-rest classifier to identify documents with strong class associations. The second tier uses a linear classifier to identify additional documents with weaker but still significant associations. This segmentation allows the system to handle large numbers of label classes while maintaining labeling completeness by capturing documents at different confidence levels across multiple classification passes.
2Measurement precision
If multiple label classes are assigned to websites, then policy filtering accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent divides the multi-label classification task into two sequential stages: first tier classification using one-vs-the-rest SVM classifiers, and second tier classification using linear SVM classifiers. This segmentation reduces computational complexity by processing documents in stages rather than requiring all classifiers to evaluate all documents simultaneously, while still achieving accurate multi-label assignment for policy filtering.
Solution Approach 2:
The patent applies partial action by implementing tiered classification where not all label classes need to be evaluated with equal depth for every document. The first tier provides a baseline classification, and the second tier adds supplementary labels only where needed. This approach achieves sufficient filtering accuracy without the full computational burden of exhaustive multi-label classification for all documents.
3Manufacturing precision
If tiered labeling approach is implemented, then multi-label assignment accuracy is improved, but processing time increases
Solution Approach 1:
The patent segments the classification process into two tiers that can be executed sequentially or in parallel. The first tier quickly identifies documents with strong class associations using optimized one-vs-the-rest classifiers. The second tier then processes documents to identify additional labels. This segmentation improves multi-label assignment accuracy while managing processing time by breaking down the complex classification task into manageable stages.
Solution Approach 2:
The first tier classification serves as preliminary action that identifies clearly classified documents before the second tier processing. Documents that are confidently classified in the first tier may not require extensive second tier analysis, allowing the system to optimize processing time by performing preliminary filtering and classification before applying more computationally intensive second tier labels where necessary.
Data Source
AI summary
The technology disclosed includes a system to perform multi-label support vector machine (SVM) classification of a document. The system creates document features representing frequencies or semantics of words in the document. Trained SVM classification parameters for a plurality of labels are applied to the document features for the document. The system determines positive and negative distances between SVM hyperplanes for the labels and the feature vector. Labels with positive distance to the feature vector are harvested. When the distribution of negative distances is characterized by a mean and standard deviation, the system further harvests the labels with a negative distance such that the harvested labels include the labels with a negative distance between the mean negative distance and zero and separated from the mean negative distance by a predetermined first number of standard deviations.