Hierarchical Graph Search Using Tensor and Fuzzy Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional search engines rely on labor-intensive hand-engineered methods and extensive synonym lists, limiting their ability to provide relevant results for semantically similar search terms and requiring significant user interaction or memory resources.
Innovation Solution
A hierarchical search method using tensor and fuzzy searches, combined with Bayesian network propagation, to determine relevance scores for nodes in a graph structure, eliminating the need for hand-engineered synonym lists and reducing memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional text-based search uses hand-engineered synonym dictionaries and fuzzy distance comparison, then search simplicity and computational scalability are achieved, but labor intensity and hand-engineering requirements increase significantly
Solution Approach 1:
The system uses unsupervised learning algorithms to automatically build synonym dictionaries and semantic relationships from the corpus itself, eliminating the need for manual curation. The model self-trains on the data to identify term relationships, making the system self-sufficient rather than requiring continuous human intervention for dictionary maintenance.
Solution Approach 2:
The patent replaces manual mechanical processes of hand-engineering synonym dictionaries with automated computational processes using machine learning models. The system substitutes human labor with algorithmic processing that automatically discovers semantic relationships through vector space modeling and similarity calculations.
2Adaptability or versatility
If extensive synonym dictionaries are hand-curated for text-based search, then semantic similarity coverage is improved, but system complexity and maintenance burden increase
Solution Approach 1:
The system changes the representation parameters from discrete synonym lists to continuous vector embeddings in a high-dimensional space. This allows semantic similarity to be captured through geometric relationships (cosine similarity, Euclidean distance) rather than explicit dictionary lookups, reducing complexity while maintaining adaptability.
Solution Approach 2:
The vector space model serves multiple functions simultaneously: it captures synonym relationships, handles misspellings through fuzzy matching, supports hierarchical categorization, and enables scalable similarity searches. This single unified approach replaces multiple specialized components that would otherwise be needed.
3Reliability
If conventional search methods rely on frequentist approaches with extensive user interaction, then search relevance is improved, but time consumption and user base requirements increase
Solution Approach 1:
The system performs preliminary actions by pre-computing vector embeddings for all terms in the corpus and pre-organizing them in the vector space before search queries arrive. This allows the search engine to immediately compute similarities without requiring iterative user interaction or frequentist sampling, delivering relevant results faster.
Solution Approach 2:
The patent introduces vector embeddings as an intermediary representation between the query and the corpus. Instead of directly comparing raw text or relying on user feedback loops, the system mediates through pre-computed semantic vectors that capture meaning, enabling faster and more reliable relevance determination.
Data Source
AI summary
Methods and systems for performing a search over hierarchical data are provided. The method may be performed by a server comprising a processor and memory. The method includes receiving a query string from a user device. The query string is searched for via a tensor search of a graph structure to determine node tensor distance score for each node in the graph structure. The query string is searched for via a fuzzy search of the graph structure to determine node fuzzy distance score for each node in the graph structure. Nodes with relevant scores are determined by updating a Bayesian network representation with evidence based on the node tensor distance scores and the node fuzzy distance scores of each node. Relevant data from the nodes is sent with relevant scores to the user device.