Document Relevancy Analysis via Cosine Distance Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning systems face challenges in providing intuitive and understandable explanations for document relevancy, as existing methods are either too complex or require significant user effort, and feature reduction algorithms often deteriorate classification accuracy in high-dimensional document classification domains.
Innovation Solution
The system calculates an internal best match score for a document by determining cosine distances between the document and training examples using term frequency-inverse document frequency weights, identifying the training example with the closest cosine distance and outputting it as an example-based explanation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex analysis algorithms are used to generate statistically valid document recommendations, then classification accuracy is improved, but user understanding and intuitiveness deteriorate
Solution Approach 1:
The patent introduces an intermediary explanation layer that bridges the gap between complex classification algorithms and user understanding. This layer generates human-readable explanations by identifying and presenting the most relevant training documents that support each classification decision, allowing users to understand the rationale without compromising the underlying complex algorithm's accuracy
Solution Approach 2:
The system creates simplified copies or representations of the complex classification process by presenting actual training documents as explanations. Instead of showing algorithmic operations, it copies the essence of the classification reasoning through relevant example documents that users can directly inspect and understand
2Ease of operation
If feature reduction algorithms are used to simplify the problem space, then user understanding is improved, but classification accuracy deteriorates
Solution Approach 1:
The patent extracts only the necessary information for explanation purposes - specifically, the most relevant training documents that justify each classification. This extraction approach provides user understanding through selected examples without removing or reducing the full feature set used by the classification algorithm, thereby maintaining accuracy while improving interpretability
3Adaptability or versatility
If interactive visualization methods are used to find explanations, then user control is improved, but computing resources and user effort increase
Solution Approach 1:
The system performs preliminary computation by pre-identifying and ranking the most relevant training documents for each test document. This preliminary action caches the explanation data in advance, reducing the need for extensive interactive computation and lowering both computing resource requirements and user effort while maintaining adaptability through on-demand retrieval
Data Source
AI summary
Systems and methods that quantify document relevance for a document relative to a training corpus and select a best match or best matches are provided herein. Methods may include generating an example-based explanation for relevancy of a document to a training corpus by executing a support vector machine classifier, the support vector machine classifier performing a centroid classification of a relevant document in a term frequency-inverse document frequency features space relative to training examples in a training corpus, and generating an example-based explanation by selecting a best match for the relevant document from the training examples based upon the centroid classification.