Hierarchical Data Grouping via Graded Similarity Membership
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data processing methods are inadequate for handling graded similarity and indiscernibility in data analytics, as they rely on crisp clustering and well-defined values, failing to efficiently group data items with multiple attributes and varying levels of granularity.
Innovation Solution
A computer-implemented method that uses graded similarity relations to partition data items into a hierarchy of classes based on predefined membership levels, allowing for the identification of similar items and triggering remedial measures in security log analysis, employing fuzzy set theory and tolerance relations to represent hierarchical categorization and indiscernibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If crisp clustering methods are used to group data items, then the grouping process is simple and well-defined, but the ability to handle graded similarity and varying levels of granularity is lost
Solution Approach 1:
The patent transforms the binary crisp clustering approach into a graded similarity approach by introducing membership levels as an additional parameter. Each data item is assigned a membership level indicating its degree of similarity to a cluster, allowing for nuanced representation of graded similarity while maintaining the fundamental clustering structure.
Solution Approach 2:
The patent segments the similarity relationship into discrete membership levels (e.g., high, medium, low similarity). This segmentation allows the system to handle graded similarity by dividing the continuous similarity spectrum into manageable categories, making the complex problem tractable while preserving the ability to represent varying degrees of similarity.
2Ease of operation
If data items are grouped at a single level of granularity, then the grouping process is straightforward, but the ability to analyze data at multiple levels of detail is lost
Solution Approach 1:
The patent adds a hierarchical dimension to the clustering structure by organizing clusters into multiple levels of granularity. Each level represents a different degree of aggregation, allowing users to navigate from coarse-grained overview clusters to fine-grained detailed clusters. This dimensional extension enables multi-level analysis while building upon the straightforward single-level clustering foundation.
Solution Approach 2:
The patent implements a nested hierarchical structure where clusters at one level contain or are related to clusters at other levels. This nesting allows data items to be grouped at multiple granularities simultaneously, with each level providing a different perspective on the data while maintaining the simplicity of individual level grouping operations.
3Productivity
If traditional clustering methods are used, then computational efficiency is maintained, but the precision of similarity measurement and detection is insufficient
Solution Approach 1:
The patent introduces membership levels as an intermediary layer between raw data items and final clusters. This intermediary structure enables more precise similarity measurement by providing intermediate classification categories, while the overall computational process remains efficient by using rule-based membership assignment rather than complex iterative algorithms.
Solution Approach 2:
The patent replaces traditional distance-based mechanical clustering mechanisms with a rule-based membership level assignment system. This substitution allows for more precise similarity measurement through configurable rules and thresholds, while maintaining computational efficiency by avoiding iterative optimization processes.
4Measurement precision
If fine-grained clustering is applied to all data, then detection accuracy is improved, but the complexity of data processing and analysis increases
Solution Approach 1:
The patent segments the data processing task by applying different levels of granularity to different data items or clusters based on their characteristics. This selective segmentation allows fine-grained processing only where necessary for accurate detection, while coarser processing is applied elsewhere, reducing overall complexity while maintaining detection accuracy where needed.
Solution Approach 2:
The patent applies fine-grained clustering with high precision only to specific regions or types of data where detection accuracy is critical, while using coarser-grained clustering for other data. This local application of quality varies the processing complexity according to the specific needs of different data portions, optimizing the balance between accuracy and complexity.
Data Source
Figure 1~2a
Figure 2b~3b
Figure 3a
AI summary
A computer implemented method of grouping a set of data items comprising the steps of: receiving a graded similarity relation between items in the set of items, each graded similarity relation defining, for a pair of items, whether the items are determined to be similar and indicating a level of similarity; receiving a definition of a set of levels of similarity as membership levels such that a pair of items being determined to have a level of similarity meeting a membership level is associated with the membership level; partitioning the set of data items into a hierarchy of classes of data item each corresponding to a membership level by determining association of data items to each membership level.