Ensemble Segmentation via Clustering and Bayesian Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Business applications often face the challenge of combining multiple competing segmentations of business data into a single unique segmentation that retains attributes from the original segmentations, without enumerating every possible combination, which is not efficiently addressed by existing methods.
Innovation Solution
A computing system with segmentation clustering and adjustment software instructions combines multiple segmentations using techniques like k-means clustering or Self-Organizing Map Neural Networks, followed by Bayesian estimation to create a final ensemble segmentation, ensuring that the information content from the original segmentations is retained.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple segmentations are combined by enumerating every possible combination, then completeness of segmentation attributes is improved, but computational complexity and time consumption increase exponentially
Solution Approach 1:
The patent applies segmentation by dividing the combination process into hierarchical levels. Instead of enumerating all possible combinations at once, the system segments the segmentation data into multiple levels (e.g., L1, L2, L3) and combines them incrementally. This allows the system to maintain attribute completeness while avoiding the exponential complexity of full enumeration by processing combinations in manageable hierarchical stages.
Solution Approach 2:
The patent implements preliminary action by pre-processing segmentation data into standardized hierarchical levels before combination. The system prepares segmentation attributes in advance with defined level structures, probability estimates, and adjustment criteria. This preliminary organization enables efficient combination operations without requiring exhaustive real-time computation during the actual segmentation merging process.
2Loss of information
If multiple segmentations are combined using traditional methods, then all segmentation attributes are retained, but the system complexity and processing requirements increase significantly
Solution Approach 1:
The patent introduces a hierarchical level dimension to organize segmentation data. By assigning segmentation attributes to specific levels (L1, L2, L3, etc.), the system transforms the complex multi-attribute combination problem into a structured hierarchical framework. This dimensional organization allows efficient processing and retention of information content while reducing system complexity through systematic level-based management of segmentation attributes.
Solution Approach 2:
The patent applies parameter changes by transforming segmentation data into standardized probability estimates and level assignments. The system converts diverse segmentation attributes into uniform parameters (probability values, level identifiers) that can be efficiently processed and combined. This parameter standardization maintains the informational content of original segmentations while simplifying the combination process through consistent parameter structures.
3Measurement precision
If probability estimates are calculated for all segment levels, then segmentation accuracy is improved, but computational resources and processing time increase
Solution Approach 1:
The patent implements partial action by calculating probability estimates selectively rather than for all possible segment combinations. The system computes probability estimates for segment levels that are actually needed based on the input segmentations, avoiding unnecessary calculations for combinations that will not be used. This selective probability estimation maintains segmentation accuracy while reducing computational resource consumption by focusing only on relevant calculations.
Data Source
AI summary
Systems and methods are provided for combining multiple segmentations into a single unique segmentation that contains attributes of the original segmentations. This new segmentation forms an ensemble or combination segmentation that has a unique set of attributes from the original segmentations without enumerating every possible set of combinations. In one example, two or more segments are combined into a single segmentation using a technique such as k-means clustering or Self-Organizing Map Neural Networks. After the first combination phase is performed, a Bayesian technique is then applied in a second phase to adjust or further alter the ensemble combination of segments.