Network Clustering via Density-Based Random Walks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network clustering methods face challenges with noisy, incomplete, and sparse labels, and struggle to handle varying densities and shapes in social networks, often resulting in large and incoherent clusters.
Innovation Solution
A density-based approach using random-walk probabilities to define node densities, allowing for adaptive clustering with user-defined thresholds to merge smaller components and control the level of supervision, enabling the discovery of clusters of varying shape and density.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional network clustering methods are used, then clustering can be performed on network data, but the resulting clusters are large and incoherent due to noisy, incomplete, and sparse labels
Solution Approach 1:
The patent transforms the clustering problem by changing the parameter representation from direct label-based clustering to density-based clustering using random-walk probabilities. This parameter transformation allows the system to work with noisy, incomplete labels by converting them into density values that capture the underlying cluster structure, thereby improving cluster coherence despite label imperfections
Solution Approach 2:
The patent introduces random-walk probabilities as an intermediary between the noisy labels and the final clustering result. This intermediary mechanism propagates label information through the network structure, allowing sparse and noisy labels to influence cluster formation indirectly through density calculations, which resolves the contradiction between having incomplete labels and achieving coherent clusters
2Adaptability or versatility
If supervised clustering with labels is applied, then clustering direction can be controlled, but the labels are noisy, incomplete, and sparse which reduces effectiveness
Solution Approach 1:
The patent changes the parameter from direct label usage to density-based representation derived from random-walk probabilities. This transformation maintains the supervisory capability of labeled clustering while making the method robust to label noise and sparsity, as the density calculation naturally smooths out label imperfections through the network structure
Solution Approach 2:
The patent enables the clustering algorithm to self-correct for noisy and sparse labels by using the network structure itself to propagate and reinforce label signals. The random-walk process automatically strengthens coherent label patterns while diminishing the impact of noisy labels, allowing the system to overcome label quality issues without external intervention
3Adaptability or versatility
If density-based clustering with random-walk probabilities is used, then adaptive clustering with varying shapes and densities is achieved, but the computational complexity increases
Solution Approach 1:
The patent makes the random-walk probability calculation serve multiple functions: it simultaneously captures cluster density, identifies cluster boundaries, and provides supervision signals. This multi-functionality reduces the need for separate computational mechanisms, thereby mitigating the increase in algorithmic complexity while achieving adaptive clustering with varying shapes and densities
Data Source
AI summary
A method (and system) for supervised network clustering includes receiving and reading node labels from a plurality of nodes on a network, as executed by a processor on a computer having access to the network, the network defined as a group of entities interconnected by links. The node labels are used to define densities associated with the nodes. Node components are extracted from the network, based on using thresholds on densities. Smaller components having a size below a user-defined threshold are merged.