Overlapping Cluster Detection via Seed Growth and Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying subsets of related items in large data sets are often inefficient and inaccurate, particularly when manual tagging is required, leading to time-consuming and inconsistent results.
Innovation Solution
A computationally efficient method for forming clusters of related items by iteratively selecting seed subsets, adding candidate items based on degrees of relatedness, and pruning items that do not meet specific criteria, allowing for overlapping clusters and automated tagging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual tagging is used to identify subsets of related items, then tagging accuracy may be improved through human judgment, but processing time increases significantly and consistency deteriorates
Solution Approach 1:
The system enables automated clustering where the data itself determines the subsets through algorithmic processing. The clustering algorithm autonomously identifies related items and forms subsets without requiring human intervention, thereby eliminating time loss while maintaining consistency through deterministic processing rules.
Solution Approach 2:
The patent replaces the manual mechanical process of human tagging with an automated computational clustering system. The clustering algorithm processes items mathematically to identify subsets, substituting human judgment with algorithmic determination, thus achieving both speed and consistency.
2Measurement precision
If traditional clustering methods are used to identify subsets, then computational accuracy may be maintained, but processing efficiency deteriorates due to high computational complexity
Solution Approach 1:
The patent segments the clustering process into distinct phases: seed selection, iterative addition of candidate items, and pruning. This segmentation allows each phase to be optimized independently, with the iterative process adding items based on local relatedness criteria and the pruning phase removing items that no longer meet thresholds, thereby improving overall efficiency while maintaining accuracy.
Solution Approach 2:
The clustering process is made dynamic through iterative addition and removal of items. The cluster composition changes over time as items are added based on current relatedness criteria and subsequently pruned if they no longer meet thresholds. This dynamic approach allows the algorithm to adapt and converge toward accurate clusters efficiently.
3Device complexity
If non-overlapping clusters are formed, then computational simplicity is maintained, but representation accuracy of real-world data deteriorates
Solution Approach 1:
The patent introduces an intermediary mechanism where items can belong to multiple clusters through overlapping memberships. This is achieved by allowing clusters to share common items, with each item's cluster assignments determined independently by the clustering algorithm. This intermediary structure enables accurate representation of real-world relationships where items naturally belong to multiple groups.
Data Source
AI summary
A technique for identifying overlapping clusters of items in a data set. The technique may be used in connection with a social network or other on-line environment in which users express approval for other users, such as through votes, tags or other inputs. These expressions of approval may be used to form clusters such that entities assigned to a cluster have a higher metric of approval from other entities within the cluster than from outside the cluster. Such clusters may be arrived at through a computationally efficient approach that involves randomly selecting one or more entities as a seed for a cluster. The cluster may be grown by testing other entities, similar to those already in the cluster, to determine whether they are more preferred by those already in the cluster than those outside the cluster. Once a cluster is grown to a desired size, it may be pruned.