Soft Correspondence Framework for Clustering Ensemble Consensus
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for combining multiple clusterings in data mining face challenges due to the correspondence problem, where different clustering algorithms or settings generate varying partitions, and existing solutions do not effectively address the correspondence issue, leading to robustness and stability issues, especially in distributed data mining and relational data clustering.
Innovation Solution
A soft correspondence framework is introduced, using a correspondence matrix to transform membership matrices into a consistent space, with an algorithm that iteratively computes consensus clustering and correspondence matrices via multiplicative updating rules, addressing the correspondence problem and providing intuitive interpretations of clustering relations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If co-association based methods are used to combine multiple clusterings, then clustering combination can be achieved, but computational complexity increases quadratically with the number of data items
Solution Approach 1:
The patent segments the correspondence problem into multiple views by introducing view-specific correspondence matrices. Instead of computing a single global correspondence matrix that requires quadratic complexity, the method divides the problem into multiple smaller sub-problems, each handling a specific view of the data. This segmentation reduces the computational burden while maintaining the ability to combine multiple clusterings effectively.
Solution Approach 2:
The patent introduces a new dimension to the problem by adding the view dimension. Instead of working with a single correspondence matrix in the original data space, the method creates correspondence matrices for multiple views, transforming the problem from a single-matrix optimization to a multi-matrix optimization across different dimensional spaces. This approach avoids the quadratic complexity of the traditional co-association method.
2Ease of operation
If re-labeling approaches are used to solve the correspondence problem, then direct correspondence can be established, but the method assumes one-to-one correspondence which limits general applicability
Solution Approach 1:
The patent makes the correspondence relationship dynamic by introducing view-specific correspondence matrices that can adapt to different data views. Instead of a static one-to-one correspondence assumption, the method allows each view to have its own correspondence matrix, enabling flexible many-to-many relationships between clusters from different clusterings. This dynamic approach maintains ease of operation through iterative optimization while significantly improving general applicability to various clustering scenarios.
3Reliability
If multiple clusterings are combined to improve robustness and stability, then better clustering solutions are obtained, but the correspondence problem becomes more difficult due to varying number and shape of clusters
Solution Approach 1:
The patent segments the correspondence problem by introducing multiple views, where each view captures different aspects of the cluster relationships. Instead of trying to solve a single complex correspondence problem with varying cluster numbers and shapes, the method divides it into multiple simpler correspondence problems, each viewed from a different perspective. This segmentation makes the correspondence detection more tractable while preserving the robustness benefits of combining multiple clusterings.
Data Source
AI summary
Combining multiple clusterings arises in various important data mining scenarios. However, finding a consensus clustering from multiple clusterings is a challenging task because there is no explicit correspondence between the classes from different clusterings. Provided is a framework based on soft correspondence to directly address the correspondence problem in combining multiple clusterings. Under this framework, an algorithm iteratively computes the consensus clustering and correspondence matrices using multiplicative updating rules. This algorithm provides a final consensus clustering as well as correspondence matrices that gives intuitive interpretation of the relations between the consensus clustering and each clustering from clustering ensembles. Extensive experimental evaluations demonstrate the effectiveness and potential of this framework as well as the algorithm for discovering a consensus clustering from multiple clusterings.