Pattern Extraction Using Frequency and Association Degree
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to effectively extract characteristic patterns from a plurality of target information pieces by considering the association between items, leading to inefficient pattern extraction and potential extraction of obvious or uninteresting patterns.
Innovation Solution
A pattern extracting apparatus that calculates an extraction evaluation value for candidate patterns based on both the frequency of appearance and association degree between items, using a candidate evaluation value calculating section to determine patterns that satisfy a predetermined threshold, thereby extracting relevant and interesting patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional frequency-based pattern extraction is used, then extraction simplicity is maintained, but pattern quality and interest are degraded
Solution Approach 1:
The patent combines frequency of appearance and association degree into a single composite evaluation metric. The candidate evaluation value calculating section integrates both factors to rank candidate patterns, merging two separate measurement dimensions into a unified evaluation framework that improves pattern quality without requiring separate extraction mechanisms.
Solution Approach 2:
The patent introduces association degree as a new parameter alongside frequency of appearance. By changing the evaluation parameters from solely frequency-based to a dual-parameter system incorporating association metrics, the system achieves more precise pattern extraction that identifies interesting patterns while maintaining manageable complexity through structured parameter integration.
2Loss of information
If association degree consideration is added to frequency-based extraction, then pattern interest is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary calculation of association degrees between item pairs before evaluating candidate patterns. The second storing section pre-stores association degree information, allowing the candidate evaluation value calculating section to quickly retrieve and apply these pre-computed values during pattern evaluation, reducing real-time computational overhead while preserving information about item relationships.
Solution Approach 2:
The patent uses frequency of appearance as a base metric and creates a weighted composite evaluation value that copies and integrates association degree information. This approach reuses existing frequency data while layering additional association-based weighting, avoiding complete recalculation from scratch and reducing computational time while preventing information loss.
3Productivity
If only frequency of appearance is used, then extraction speed is maintained, but pattern characteristic quality deteriorates
Solution Approach 1:
The candidate evaluation value calculating section automatically integrates both frequency and association degree metrics without requiring external intervention or manual pattern filtering. The system self-evaluates candidate patterns using the composite metric, maintaining extraction speed through automated evaluation while improving pattern quality by considering association characteristics inherent in the data.
Solution Approach 2:
The patent changes the evaluation parameter from single-factor frequency to a composite parameter incorporating both frequency and association degree. This parameter transformation enables the system to maintain extraction speed through automated calculation while significantly improving pattern characteristic quality by identifying patterns with meaningful item associations rather than merely frequent co-occurrences.
Data Source
AI summary
A pattern extracting apparatus includes a first storing section storing plural target information pieces, a candidate pattern producing section producing candidate patterns each including two or more items different from each other based on each of the items included in the plural target information pieces, a candidate evaluation value calculating section calculating an extraction evaluation value of the candidate pattern based on a frequency of appearance at which the produced candidate pattern appears in the plural target information pieces, a pattern extracting section determining and extracting any of the candidate patterns having the calculated extraction evaluation value satisfying a predetermined threshold value, and a second storing section storing an association degree between the items. The candidate evaluation value calculating section calculates the extraction evaluation value based on a weight based on an identified association degree between the items included in the candidate pattern and the frequency of appearance.