Latent Keyword Generation via Cohesion and Context Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for selecting keywords from documents only choose phrases that directly occur in the document, failing to identify latent keywords that do not appear in the document.
Innovation Solution
A method and apparatus that generate latent keywords by combining primitive single words, calculating cohesion scores based on co-occurrence within a predetermined window range, and context scores based on similarity between contexts in the target and neighboring documents, allowing for the selection of keywords that do not directly occur in the document.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If candidate keywords are extracted only from phrases that directly occur in the document, then the extraction process is simple and reliable, but latent keywords that do not occur in the document cannot be selected
Solution Approach 1:
The patent segments keywords into two categories: explicit keywords that directly occur in the document and latent keywords generated from combinations of primitive single words. This segmentation allows the system to handle both types of keywords through different processing paths, expanding the selection range while managing complexity through modular design.
Solution Approach 2:
The patent performs preliminary actions by extracting primitive single words from the document before generating latent keywords. By preparing these basic building blocks in advance, the system can efficiently generate candidate latent keywords through combination without excessive computational complexity during the main keyword selection process.
2Adaptability or versatility
If latent keywords are generated by combining primitive single words, then the range of selectable keywords is expanded, but the complexity of calculating cohesion and context scores increases
Solution Approach 1:
The patent introduces intermediary scoring mechanisms (cohesion score and context score) that mediate between the generated latent keywords and the final keyword selection. These intermediary scores break down the complex evaluation into manageable components: cohesion score measures word pairing strength while context score evaluates contextual relevance, making the overall complexity tractable.
Solution Approach 2:
The patent changes parameters by introducing multiple scoring dimensions (cohesion score based on co-occurrence frequency, context score based on contextual similarity) rather than relying on a single complex score. This parameterization allows the system to evaluate latent keywords through multiple independent metrics that can be calculated and combined systematically.
3Loss of information
If only explicit keywords are selected, then the keyword selection process is straightforward and reliable, but the system cannot identify concepts that are implied but not explicitly stated in the document
Solution Approach 1:
The patent creates copies of primitive single words through systematic combination to generate latent keywords that represent implied concepts. Instead of directly extracting latent concepts from the document, the system copies and recombines existing words to construct new keyword candidates that capture implicit meanings, thereby reducing information loss without requiring complex direct inference mechanisms.
Data Source
AI summary
Provided are a method and an apparatus for generating a keyword. The method for generating a keyword in a target document includes: extracting primitive single words from the target document; generating a candidate keyword by a combination of the primitive single words; calculating cohesion score between a first word and a second word included in the candidate keyword; calculating a context score of the first word and the second word based on similarity between two contexts by determining a periphery of a word where the first word and the second word simultaneously occur in a neighboring document as one context and determining the target document as a remaining one context; and selecting a final keyword based on the cohesion score between the first word and the second word and the context score of the first word and the second word.