Paragraph-Based Document Search Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional search engines fail to effectively distinguish between relevant and tangentially-related sections in documents, leading to inefficiencies in retrieving relevant legal and other types of documents, especially when search queries are long or contain non-central terms.
Innovation Solution
A method that generates paragraph scores by associating term weight values with paragraph terms, comparing these scores to search terms, and combining them to determine overall document scores for improved document retrieval, using standardized paragraph terms and search terms to enhance search accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional search engines rank documents based on distance between matched words, then search speed is improved, but search accuracy deteriorates because documents with tangentially-related sections are incorrectly ranked highly
Solution Approach 1:
The patent segments documents into paragraphs and calculates search scores at the paragraph level rather than document level. Each paragraph is independently scored based on term matching, allowing precise identification of relevant sections. This segmentation enables the system to distinguish between documents with truly relevant content versus those with only tangential mentions, resolving the accuracy issue while maintaining efficiency through localized scoring.
Solution Approach 2:
The patent applies local quality by assigning different weights to different paragraphs within a document based on their relevance to the search query. Instead of treating all paragraphs equally, the system calculates individual paragraph scores and combines them weighted by relevance. This allows highly relevant paragraphs to contribute more to the overall document score, improving search accuracy without sacrificing speed.
2Productivity
If search terms appear in multiple sections of a document, then the document's rank increases in conventional search, but relevance accuracy deteriorates because non-central sub-topics inflate the ranking
Solution Approach 1:
By segmenting the document into paragraphs and scoring each independently, the patent prevents non-central sub-topics from inflating the overall document rank. Only paragraphs containing central, relevant information receive high scores. This segmentation approach maintains productivity by efficiently processing multiple sections while improving relevance accuracy through selective scoring of only the truly relevant paragraphs.
Solution Approach 2:
The patent applies partial action by focusing scoring efforts only on the most relevant paragraphs rather than uniformly processing all sections. The system identifies and scores key paragraphs that contain the central thesis or main arguments, allowing non-essential sections to contribute minimally or not at all to the final ranking. This selective approach improves relevance accuracy while maintaining productivity.
3Reliability
If users must manually review search results to identify relevant documents, then comprehensive search coverage is improved, but time consumption increases
Solution Approach 1:
The patent performs preliminary action by pre-calculating paragraph scores and organizing documents based on these scores before user queries are submitted. The system maintains an indexed structure of paragraph-level relevance metrics, allowing rapid retrieval and ranking of relevant documents in response to user queries. This preliminary organization ensures comprehensive search coverage while minimizing user time consumption by presenting pre-ranked results.
Solution Approach 2:
The patent replaces the mechanical manual review process with an automated paragraph-based scoring and ranking system. The system automatically calculates relevance scores, ranks documents, and presents results in order of relevance. This substitution maintains search completeness through thorough paragraph-level analysis while dramatically reducing user time consumption by eliminating the need for manual evaluation of each search result.
Data Source
AI summary
Computer programs for searching a collection of electronic documents are disclosed. A computer program for searching a collection of electronic documents, includes instructions that cause the computer to carry out steps of: replacing a set of synonymous terms appearing in a paragraph with a set of standardized paragraph terms, where each standardized paragraph term has an associated term weight and generating standardized search terms in response to a search query, generating paragraph scores for paragraphs of a document based at least in part on the associated weights of standardized paragraph terms that match one or more of the standardized search terms. The computer program further includes determining overall document scores for the electronic documents based at least in part on a combination of the paragraph scores, and determining a set of matching documents, wherein the set of matching documents is ordered using the overall document scores.