Text Categorization Using External Knowledge Base
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computerized text categorization methods face limitations due to the lack of external knowledge, leading to performance barriers in accurately classifying documents, as they fail to utilize contextual information and resolve issues like polysemy and synonymy effectively.
Innovation Solution
A system and method that utilizes a computerized encyclopedia to provide weighted concepts related to input text, leveraging external knowledge to enhance feature vectors and improve categorization by incorporating concepts from a knowledge database, which includes documents explaining each concept, and uses these to generate a classifier that determines the most relevant categories and documents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional bag-of-words methods are used for text categorization, then the system is simple and easy to implement, but the categorization accuracy is limited due to lack of external knowledge
Solution Approach 1:
The patent introduces an external knowledge base as an intermediary component between the input text and the categorization engine. This knowledge base stores pre-acquired knowledge about entities, concepts, and their relationships, which mediates the categorization process by providing contextual information that the traditional bag-of-words method lacks, thereby improving accuracy without significantly increasing system complexity
Solution Approach 2:
The system performs preliminary action by pre-acquiring and storing knowledge in the knowledge base before the actual categorization task. This pre-processing step involves collecting information about entities, concepts, and their relationships, which is then readily available during categorization to improve accuracy without adding complexity to the core categorization algorithm
2Measurement precision
If external knowledge from a computerized encyclopedia is integrated, then the ability to resolve polysemy and synonymy improves, but the processing time and computational resources increase
Solution Approach 1:
The system performs preliminary action by pre-acquiring and storing knowledge about entities, concepts, and their relationships in a knowledge base before the actual categorization task. This pre-processing step enables fast lookup during categorization, resolving polysemy and synonymy without adding significant processing time to the main workflow
Solution Approach 2:
The patent applies local quality by selectively applying external knowledge enhancement only to specific portions of the text where ambiguity or synonymy is detected, rather than processing the entire text uniformly. This targeted approach improves word sense disambiguation accuracy while minimizing the additional computational overhead
3Measurement precision
If feature vectors are enhanced with concepts from the knowledge base, then the categorization performance improves, but the dimensionality and complexity of the feature space increases
Solution Approach 1:
The patent applies universality by using the same knowledge base to serve multiple functions: it provides entity recognition, concept enrichment, and contextual information for categorization. This multi-functional use of a single knowledge structure improves categorization performance without proportionally increasing feature space complexity, as the same knowledge resources support multiple enhancement tasks
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of providing weighted concepts related to a sequence of one or more words, including: providing on a computer an encyclopedia with concepts and a document explaining each concept, forming a vector, which contains the frequency of the word for each concept, for each word in the encyclopedia, arranging the vector according to the frequency of appearance of the word for each concept, selecting the concepts with the highest frequencies for each word from the vector, truncating the rest of the vector, inducing a feature generator using the truncated vectors; wherein the feature generator is adapted to receive as input one or more words and provide a list of weighted concepts, which are most related to the one or more words provided as input.