Data Classification Ontology Trees Small Dataset Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data classification algorithms perform poorly on small datasets due to imprecise boundaries between classes, leading to incorrect predictions and reduced classification accuracy, as they do not capture sufficient variance and often classify new data into multiple categories.
Innovation Solution
The method involves creating category ontology trees from a small dataset, extracting input keywords, computing membership scores based on these trees, and determining the classification of an input statement using a scoring system that considers the depth and reachability of keywords within the trees.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional classification algorithms (LSI, SVM) are used on small datasets, then the algorithms can process the data, but the classification accuracy deteriorates due to insufficient data variance and imprecise class boundaries
Solution Approach 1:
The patent segments the classification problem into multiple hierarchical levels using category ontology trees. Instead of treating all classes uniformly, the system divides them into parent categories and sub-categories, allowing progressive classification at different granularities. This segmentation enables accurate classification even with limited data by focusing variance detection at appropriate hierarchical levels.
Solution Approach 2:
The patent introduces a new dimensional approach by computing membership scores that quantify the degree of belonging to each category. Instead of binary classification, the system adds a scoring dimension that measures confidence and ambiguity, enabling more nuanced classification decisions when data is limited. The membership score dimension captures the uncertainty inherent in small datasets.
2Loss of time
If machine learning algorithms are trained on small datasets, then training time is reduced, but the boundaries between classes become imprecise leading to incorrect predictions
Solution Approach 1:
The patent performs preliminary action by pre-computing membership scores for training data points and storing them in the category ontology trees. This pre-computation captures the essential variance patterns in advance, allowing the system to make accurate predictions without re-processing the entire dataset during inference. The preliminary scoring establishes reliable class boundaries even with limited training data.
Solution Approach 2:
The system incorporates feedback mechanisms where membership scores are computed and used to refine classification decisions. The scoring system provides continuous feedback about the confidence level of predictions, allowing the system to identify cases where additional analysis is needed. This feedback loop improves reliability by flagging uncertain predictions that may require manual review or additional data.
3Productivity
If classification algorithms are applied to small datasets, then processing speed is maintained, but multiple categories are incorrectly assigned to single input statements due to overlapping boundaries
Solution Approach 1:
The patent changes the parameter representation by using membership scores instead of binary category assignments. This parameter transformation allows the system to express classification confidence and ambiguity quantitatively. By changing from discrete category labels to continuous membership scores, the system can distinguish between clear and ambiguous classifications, reducing misclassification while maintaining processing efficiency.
Data Source
AI summary
This technology relates generally to data classification, and more particularly to system and method for classifying an input statement with respect to a small dataset. In one embodiment, a method is provided for classifying an input statement with respect to a small dataset. The method comprises receiving the input statement, extracting a plurality of input keywords from the input statement, computing a membership score for each of a plurality of category ontology trees based on the plurality of input keywords, and determining a classification of the input statement based on the membership score. The plurality of category ontology trees are derived from the small dataset and correspond to a plurality of classifications in the small dataset.