Hierarchical Document Clustering via User Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for constructing a tree of clusters of electronic documents either require manual effort, which is tedious for large datasets, or automatic classification that does not align with user preferences and lacks flexibility for modification.
Innovation Solution
A method that allows users to define constraints between documents, enabling the construction of a hierarchical document cluster tree with minimal interaction, where users can modify the classification preferences and structure dynamically by obtaining, storing, and iterating on constraints, and automatically reorganizing the tree based on user requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If manual classification method is used, then the tree structure conforms to user wishes and facilitates manual modification, but it becomes very tedious when there are many documents to be listed
Solution Approach 1:
The patent segments the classification process into hierarchical levels, where users define constraints at each level rather than manually placing every document. The system divides documents into clusters at different tree levels based on user-defined constraints, reducing the overall complexity and time required for classification while maintaining user control over the structure.
Solution Approach 2:
The patent applies preliminary action by having users define constraints and preferences before the actual classification process. Users specify must-link and cannot-link constraints, similarity thresholds, and other parameters in advance, allowing the system to automatically perform the time-consuming document clustering while respecting user-defined structural requirements.
2Productivity
If totally automatic classification method is used, then classification speed is fast, but the result does not always correspond to the organization required by the user and no correction is possible
Solution Approach 1:
The patent implements feedback by allowing users to review the automatically generated tree structure and provide corrections through constraint definitions. Users can specify must-link constraints for documents that should be together, cannot-link constraints for documents that should be separated, and the system iteratively adjusts the clustering to satisfy these feedback constraints while maintaining overall automation.
Solution Approach 2:
The patent makes the classification system dynamic by allowing users to modify constraints and re-run classification at any level of the tree. The system can dynamically adjust cluster assignments based on user feedback without requiring complete reclassification, enabling flexible adaptation to user organizational requirements while maintaining fast processing speeds.
3Manufacturing precision
If supervised classification method is used, then documents are classified according to user-defined labels, but users must know all possible labels a priori which greatly restricts its use
Solution Approach 1:
The patent inverts the traditional supervised classification approach by not requiring users to predefine all possible labels or cluster structures. Instead of users specifying what clusters should exist, the system automatically discovers the hierarchical structure based on user-defined constraints and document similarities, eliminating the need for a priori knowledge of the classification scheme.
Solution Approach 2:
The patent changes the parameters users need to define from fixed labels and cluster assignments to flexible constraints (must-link, cannot-link, similarity thresholds). This parameter transformation allows users to express organizational preferences without needing to know the complete classification structure in advance, reducing complexity while maintaining classification accuracy.
4Adaptability or versatility
If semi-supervised classification with constraints is used, then user preferences are taken into account, but the method works only for non-hierarchical classifications and offers no solution for modifying the classification obtained
Solution Approach 1:
The patent implements a nested hierarchical structure where clusters at lower levels are contained within clusters at higher levels. Users can define constraints at multiple hierarchical levels, and the system maintains the nested structure while allowing modifications at any level. This enables both hierarchical organization and flexible modification by allowing users to add, remove, or adjust constraints at specific levels without affecting the entire classification structure.
Data Source
AI summary
A method of assisting with the construction of a tree of clusters of electronic documents, the documents being defined by predetermined attributes. The method includes, for a given cluster of documents and a given level of the tree, the following steps: a) obtaining (E300) constraints defined between at least two documents of said cluster and storing those constraints; b) constructing (E310) subclusters as a function of the constraints obtained and the attributes of the documents of said cluster; c) creating (E320) a tree level corresponding to the subclusters created; and d) selecting (E340) a created subcluster and iterating the steps a), b), c) and d) for the selected subcluster until said tree is obtained. A device for implementing the method is also disclosed.