Document Classification Using Monotonic Regression Score Combination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automatic classifiers often lack the confidence level needed for reliable classification of documents, especially when classifying entire websites, as they provide insufficient likelihood scores, making it difficult to take automatic actions and requiring high confidence for site-level actions, which is harder to achieve due to limited information.
Innovation Solution
The method involves selecting multiple classifiers to score documents based on a specified property, combining scores using a multiple classifier model that employs monotonic regression, allowing for high-confidence classification and filtering of search results by generating a probability model that can classify documents without human evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional automatic classifiers are used to classify documents, then classification speed is improved, but classification confidence and reliability deteriorate
Solution Approach 1:
The patent combines multiple independent classifiers (text-based classifier, image-based classifier, URL-based classifier) into a unified classification system. Each classifier processes different aspects of the document independently, and their results are merged through a meta-classifier that aggregates the individual classification scores to produce a final classification decision, thereby improving reliability while maintaining automated processing speed
Solution Approach 2:
The classification system uses a composite approach by integrating multiple classification methods (text analysis, image analysis, URL analysis) into a single unified framework. The meta-classifier combines the outputs of heterogeneous classifiers with different strengths, creating a composite classification system that leverages the advantages of each individual classifier while compensating for their individual weaknesses
2Reliability
If site-level classification is performed with high confidence requirements, then action reliability is improved, but classification applicability deteriorates
Solution Approach 1:
The patent segments the classification task into two distinct levels: site-level classification (applying to all pages of a website) and page-level classification (applying to individual pages). The system first attempts site-level classification using aggregate features, and only when confidence thresholds are not met does it fall back to page-level classification, allowing high reliability requirements to be satisfied while maintaining versatility through the hierarchical approach
Solution Approach 2:
The system performs preliminary site-level classification before proceeding to page-level classification. By first evaluating the entire site using aggregate features and classification scores, the system can quickly resolve many cases at the higher confidence level, and only performs the more computationally intensive page-level analysis when necessary, thereby maintaining both reliability and applicability
3Adaptability or versatility
If page-level classification is performed with limited information, then classification versatility is improved, but measurement precision deteriorates
Solution Approach 1:
The classification system is designed to operate universally at both site-level and page-level, adapting its behavior based on the available information and confidence requirements. The same meta-classifier framework and multiple base classifiers (text, image, URL) are used in both modes, allowing the system to maintain measurement precision by selecting the appropriate classification level while preserving versatility through its ability to switch between site-level and page-level operation
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for classifying resources using scores from multiple classifiers. In general, one aspect of the subject matter described in this specification can be embodied in methods that include the actions of receiving identifying a collection of documents to classify; receiving a plurality of classifiers for scoring a document with respect to a specified property; for each document in the collection, applying each of the plurality of classifiers, each classifier generating a score associated with a likelihood that the document has the specified property, combining the scores from each classifier including applying a multiple classifier model that uses monotonic regression to combine the plurality of classifiers, and classifying the document as having the specified property based on the combined score.