Hierarchical Similarity Index Labeling for Subset Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for similarity-based calculations, such as those using ε-k-d trees and classification search trees, face performance issues when calculating similarity for subsets of data or in distributed environments, as they require costly integration or reconstruction of similarity indices.
Innovation Solution
A device and method that store a similarity index with hierarchically configured nodes representing similarity relations, generate labels for data based on node positions, and reconfigure the index using these labels for targeted similarity calculations, allowing efficient calculations on subsets of data without full index reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If local similarity indices are integrated into one to perform similarity-based calculation for any subset, then versatility is improved, but device complexity and processing cost increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-storing the hierarchical structure of the similarity index (including node relationships, depth information, and range data) before subset queries are requested. This allows the system to quickly reconfigure and extract relevant portions of the index for any subset without performing costly integration operations, thereby resolving the contradiction between versatility and complexity.
Solution Approach 2:
The patent segments the similarity index into a hierarchical tree structure where data is organized by depth levels and ranges. Each node contains information about its position and the data range it covers. This segmentation enables efficient extraction of subset data by selectively traversing and reconfiguring only the relevant portions of the hierarchy, avoiding the need to integrate entire local indices.
2Measurement precision
If similarity index is reconstructed for any subset, then calculation accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by reconstructing only the necessary portion of the similarity index required for the specific subset query, rather than rebuilding the entire index. The system determines the minimal subtree or node range needed based on the subset parameters and performs reconstruction only for that portion, thereby maintaining accuracy while significantly reducing processing time.
Solution Approach 2:
The patent stores pre-computed metadata about the hierarchical index structure including node depths, ranges, and relationships. This preliminary preparation allows the system to quickly identify which portions of the index need reconstruction for any given subset, avoiding full index rebuilding and reducing processing time while maintaining calculation accuracy.
3Productivity
If data is divided into sub data sets for parallel processing, then productivity is improved, but difficulty of detecting and measuring similarity across sub data sets increases
Solution Approach 1:
The patent creates a universal hierarchical similarity index structure that can serve multiple sub-data sets simultaneously. The index uses standardized node formats, depth levels, and range representations that work consistently across all subsets. This universality allows parallel processing of different subsets while maintaining the ability to detect and measure similarities across subset boundaries through the common hierarchical framework.
Solution Approach 2:
The patent introduces a hierarchical dimension to the data organization, arranging data in tree structures with multiple levels of abstraction. This dimensional transformation allows parallel processing at different hierarchy levels, where leaf nodes represent individual data points and parent nodes represent aggregated groups. The hierarchical dimension enables efficient cross-subset similarity detection by providing a common framework that spans all subsets.
Data Source
AI summary
A similarity index storage unit stores a similarity index in which nodes storing attribute values related to similarities among data are hierarchically arranged so as to indicate similarity relationships among the data. A label generation unit generates a label for each data, said label being based on paths which indicate the positions of the nodes storing the attribute values related to similarities in the similarity index. A similarity calculation unit reconfigures, on the basis of the label, the similarity index of data indicated by all or a portion of information associating the attribute values related to similarities with the label, and executes similarity-based calculation of the all or partial data by using the reconfigured similarity index.