Object Clustering via Weighted Directed Network Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object clustering methods suffer from low precision due to clustering based solely on identifiers of association objects, leading to incorrect grouping of target objects, where objects with strong associations are not clustered together and those with weak associations are incorrectly grouped.
Innovation Solution
The method constructs a directed network graph with weighted edges representing the similarity between target objects, iteratively reassigned category identifiers based on in-degree node groups with the largest total weight, and clusters objects based on these identifiers to improve clustering precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clustering is performed based solely on identifiers of association objects, then the clustering process is simple and fast, but the clustering precision is low
Solution Approach 1:
The patent transforms the clustering problem from a one-dimensional identifier matching task into a multi-dimensional weighted graph problem. By constructing a directed network graph where nodes represent target objects and edges represent association relationships with weights indicating similarity degrees, the system captures complex relationships beyond simple identifier equality. This dimensional transformation enables precise clustering by considering multiple association dimensions simultaneously.
Solution Approach 2:
The patent introduces weight parameters to quantify the similarity degrees between association objects, transforming the binary matching problem into a continuous optimization problem. By calculating weights based on multiple factors (such as the number and strength of associations) and using these weights to guide clustering decisions, the system achieves higher precision. The iterative reassignment of category identifiers based on weighted in-degree node groups further refines the clustering results.
2Reliability
If traditional clustering methods are used, then objects with the same association object identifier are grouped together, but objects with strong associations may not be clustered together and objects with weak associations may be incorrectly grouped
Solution Approach 1:
The patent implements an iterative feedback mechanism where category identifiers are repeatedly reassigned based on the weighted in-degree node groups. In each iteration, the system evaluates the current clustering state, identifies in-degree nodes with the largest total weights, and reassigns category identifiers to improve clustering accuracy. This feedback loop continues until convergence, ensuring that objects with strong associations are correctly grouped while minimizing incorrect groupings.
Solution Approach 2:
The patent replaces the simple mechanical identifier-matching mechanism with a sophisticated weighted graph analysis system. Instead of directly grouping objects with identical association identifiers, the system constructs a directed network graph, calculates weighted edges representing similarity degrees, and uses iterative optimization algorithms to determine final clustering. This substitution of mechanical operations with computational analysis significantly improves clustering reliability.
Data Source
AI summary
System and methods for object clustering are provided. The system may determine a weight of a directed edge between nodes of a directed network graph. The weight may represent a similarity degree of target objects associated with the target nodes. The similarity degree may reflect strength of an association between the two target objects, and further reflect a possibility that the two target objects belong to the same clustering category. After the directed network graph is constructed, nodes in the directed network graph may be clustered based on weights of directed edges between the nodes in the directed network graph.