Knowledge Graph Keyword Search Using Structural Compact Subgraphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for keyword search in knowledge graphs are computationally demanding and often return empty results when input entities are distantly distributed, as they rely on group Steiner trees which are NP-hard and have high run times, especially for large graphs, and structurally compact subgraphs are not guaranteed, limiting their applicability.
Innovation Solution
A method that computes a salient subset of a knowledge graph to find structural compact subgraphs connecting input entities within a predefined diameter bound, using algorithms that prioritize entity salience and distance computation to avoid expensive searches and ensure structural compactness, merging paths with common end vertices to form compact subgraphs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If group Steiner tree approach is used to find relationship subgraphs, then the subgraphs connect input entities with minimum weight, but the computation time becomes prohibitively high for large knowledge graphs
Solution Approach 1:
The patent changes the optimization parameter from minimum weight (Steiner tree) to structural compactness constraints (diameter bound, order bound). This transforms the NP-hard problem into a tractable problem by imposing predefined structural constraints that limit the search space while still ensuring connectivity and compactness of the relationship subgraphs.
Solution Approach 2:
The patent segments the knowledge graph search space by imposing structural constraints (diameter bound, order bound) that divide the potentially infinite search space into manageable bounded regions. This segmentation allows efficient exploration of only those subgraphs that satisfy the compactness constraints, avoiding exhaustive search of the entire graph.
2Productivity
If structurally compact subgraphs with predefined constraints are used, then computation efficiency and user perception are improved, but empty results are returned when input entities are distantly distributed
Solution Approach 1:
The patent introduces dynamic adaptability by allowing the system to switch between different search strategies based on the distribution of input entities. When entities are closely distributed, structurally compact subgraphs are used; when distantly distributed, the system can relax constraints or use alternative approaches, making the method adaptive to different scenarios rather than rigidly applying fixed constraints.
Solution Approach 2:
The patent introduces intermediate structures (such as virtual entities or relay nodes) that can serve as mediators to connect distantly distributed input entities while maintaining structural compactness. These intermediary elements allow the construction of compact relationship subgraphs even when direct connections between input entities would violate diameter or order bounds.
3Manufacturing precision
If minimum weight Steiner trees are computed, then optimal connection is achieved, but the subgraphs may be too large for user perception
Solution Approach 1:
The patent changes the optimization focus from weight minimization to structural compactness by imposing diameter bounds and order bounds. This parameter change ensures that the relationship subgraphs remain visually manageable and comprehensible to users, with limited depth and breadth, while still providing meaningful connections between input entities.
Solution Approach 2:
The patent applies preliminary structural constraints (diameter bound, order bound) before performing the actual subgraph computation. This preliminary action pre-defines the maximum acceptable size and shape of the result, ensuring that the computed subgraphs will be compact and suitable for user perception from the outset, rather than requiring post-processing to reduce size.
Data Source
AI summary
A computer-implemented method for keyword search in a data set. Data of the data set is represented by a knowledge graph. The knowledge graph comprises vertices representing entities of the data set and edges representing relations between said entities. The method comprises the following steps: receiving a search query comprising at least two entities; computing for the at least two entities of the search query a salient subset of the data set, wherein the salient subset is computed such that a structural compact subgraph exists in the knowledge graph, the structural compact subgraph connecting the at least two entities of the search query, and computing for the salient subset a structural compact subgraph of the knowledge graph which connects the at least two entities of the search query.