Semantic Similarity Scoring for Text Classification Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text classifiers in search engines often retrieve large volumes of documents, requiring users to manually inspect results to find specific information, as they rely on word matching and do not effectively utilize semantic similarity for improved precision and recall.
Innovation Solution
A computer-based method that expands principal words into sets of semantically similar words, calculating membership degrees and using these scores to rank search results, thereby enhancing the accuracy and efficiency of document retrieval by reducing ambiguity and improving keyword specificity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If word matching processes are used for text classification, then the search system is simple to implement, but the precision of information retrieval deteriorates due to retrieval of tens of thousands to millions of documents requiring manual inspection
Solution Approach 1:
The patent changes the parameter of word representation from exact word matching to semantic similarity measurement. By calculating degrees of membership that reflect semantic similarity in meaning, the system transforms rigid word matching into a flexible semantic comparison, thereby improving retrieval precision without excessive complexity increase
Solution Approach 2:
The patent introduces sets of semantically similar words as an intermediary layer between the user's search query and the document corpus. This intermediary layer with calculated membership degrees acts as a bridge that captures semantic intent, improving precision while maintaining system manageability
2Measurement precision
If sets of semantically similar words are used for text classification, then the precision and recall of text classifiers is improved, but the complexity of the search system increases due to calculating degrees of membership and computing scores
Solution Approach 1:
The patent performs preliminary action by pre-determining sets of semantically similar words and their membership degrees before the actual search execution. This preparation work is done in advance, allowing the search system to utilize pre-computed semantic relationships, thereby reducing the complexity burden during runtime while maintaining improved precision and recall
Solution Approach 2:
The patent segments the search process into distinct components: determining semantically similar words, calculating membership degrees, performing the search, computing scores, and sorting results. This segmentation allows each component to be optimized independently, managing overall system complexity while achieving superior classification performance
3Loss of time
If traditional keyword-based retrieval is used, then the ease of operation is maintained, but the loss of time increases due to users having to manually inspect retrieved documents
Solution Approach 1:
The patent implements feedback by computing scores based on degrees of membership and using these scores to sort and rank search results. This feedback mechanism automatically prioritizes the most relevant documents at the top of the results list, significantly reducing the time users need to spend manually inspecting documents while maintaining ease of operation through automatic ranking
Data Source
AI summary
A computer-based searching method is described that includes receiving, at a computer, a search concept that includes a plurality of principal words, determining a set of semantically similar words for at least one of the principal words, calculating a degree of membership for each word in each set that reflects a semantic similarity in meaning to the principal word of the set, performing the search, computing a score based at least on the degrees of membership for the words in the word set, and sorting the results using the computed scores.