Rule-Based Text Classification Using Category Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated text classification methods rely on statistical characteristics of language, which are inadequate for specialized domains where each word has specific meanings, leading to multiple contexts and inadequate classification.
Innovation Solution
A rule-based classification system that maps words from text fragments to category vectors using specific rule functions, generating a candidate classification decision matrix to classify text fragments based on contextual patterns rather than probabilistic language characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If statistical characteristics of language are used for text classification, then generic classification can be achieved, but classification accuracy in specialized domains deteriorates
Solution Approach 1:
The patent applies local quality by using domain-specific category map tables instead of generic statistical models. Each domain (e.g., medical, legal, technical) has its own customized category mappings that reflect local terminology and contextual relationships, allowing the system to adapt the classification approach to the specific requirements of each domain while maintaining overall system versatility
Solution Approach 2:
The system changes parameters by switching between different category map tables based on the domain of the text being classified. Rather than using fixed statistical characteristics, the system dynamically adjusts the classification parameters (category mappings, rule sets) to match the specific domain, thereby improving accuracy while maintaining adaptability across different domains
2Measurement precision
If domain-specific rule functions are applied, then classification accuracy in specialized domains is improved, but system complexity increases
Solution Approach 1:
The patent segments the classification system into modular components: domain-specific category map tables, rule functions, and decision logic. Each domain can be independently configured with its own category mappings and rules, allowing complex domain-specific classification to be broken down into manageable, reusable modules that can be maintained and updated separately
Solution Approach 2:
The system achieves universality through a standardized framework that can handle multiple domains. The same basic architecture (category map tables, rule functions, decision logic) serves all domains, reducing overall complexity by avoiding the need for separate classification systems for each domain while still providing domain-specific accuracy
Data Source
AI summary
A method for classifying a text fragment by applying one or more rule functions is provided. The method includes (i) obtaining a set of words from one or more sentences of the text fragment obtained from a multimedia content, (ii) mapping each word from the set of words with one or more category map tables to obtain a set of candidate vector, each candidate vector includes at least one category associated with each word, (iii) generating one or more category vectors based on the at least one category, (iv) applying rule functions on the one or more category vectors, (v) generating a candidate classification decision matrix based on the rule functions, and (vi) classifying the text fragment based on the candidate classification decision matrix. The candidate classification matrix includes results of each rule function applied on each of the category vector.