Multi-way nested search engine with concept tagging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current information retrieval systems face challenges in delivering highly relevant search results due to issues like synonymy, polysemy, spelling errors, abbreviations, and partial matches, and struggle with whole document searches that require handling discrepancies in both query and document terms, lacking precision and incorporating human knowledge effectively.
Innovation Solution
The multi-way nested searching technique employs contextual personalized search methods using canonical semantic knowledge represented in knowledge bases, allowing for reference, reverse, and wish list searches to be combined, enabling a search engine to score and rank documents based on similarity, thereby producing the most relevant results by leveraging graph relationships between concepts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If keyword searching or Boolean queries are used, then the search system is simple to implement, but it cannot address synonymy, polysemy, spelling errors, abbreviations, and partial matches
Solution Approach 1:
The patent introduces concept tagging as an intermediary layer between keywords and documents. Concepts serve as mediators that connect related terms (synonyms, abbreviations, spellings) to their underlying meaning, allowing the system to handle linguistic variations while maintaining implementation feasibility through structured concept hierarchies
2Reliability
If concept tagging is used to map documents and queries into concept space, then synonymy and polysemy can be addressed, but it is difficult to accurately extract concepts from documents with high precision and recall
Solution Approach 1:
The patent performs preliminary concept extraction and tagging during document indexing before the actual search occurs. By pre-processing documents to assign concept tags and hierarchies in advance, the system reduces the complexity of real-time concept extraction while maintaining high precision through established taxonomic structures
Solution Approach 2:
The system incorporates feedback mechanisms where search results and user interactions inform concept extraction improvements. Concept hierarchies and taxonomies are refined based on search performance data, allowing the system to learn from actual usage patterns and improve concept extraction accuracy over time
3Adaptability or versatility
If automatic classification/categorization algorithms are used, then documents can be organized into taxonomies, but they commonly have low accuracy, poor results, and require significant training
Solution Approach 1:
The patent segments the classification task into hierarchical levels, organizing concepts into parent-child relationships within taxonomies. This segmentation allows documents to be classified at multiple granularity levels simultaneously, improving accuracy by breaking down complex classification into manageable hierarchical steps rather than requiring single-pass accurate classification
4Reliability
If latent semantic indexing is used to map words into reduced concept space, then similarity measures can be produced, but it does not allow high precision in correlation between concepts nor incorporate human knowledge
Solution Approach 1:
The patent introduces structured concept taxonomies as intermediaries between raw text and similarity measurement. Instead of directly comparing words in reduced space, the system maps both query and document terms through concept hierarchies that incorporate human knowledge and semantic relationships, enabling more precise correlation measurements that reflect actual meaning rather than just statistical co-occurrence
Data Source
AI summary
A multi-way nested search utilizes one or more knowledge bases to allow a search engine to search through a database of target documents to find the best match for a given situation. In one node of the search, a user can input one or more search terms in an input string that can be searched against the database of target documents to select and score a resulting set of documents. In another node, a reference document can be input and search terms extracted for searching against the database of target documents to select and score a resulting set. In yet another node, a reverse search can be conducted in which the reference document is scored against the result set of target documents from the reference document search. These various search results and scores can be combined using combining functions to ultimately produce the most relevant set of target documents for the user. The user can control how much weigh is given to each search.