Natural Language Query Answering via Sentence Scoring and Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Search engines face challenges in providing natural language answers to natural language questions, as they typically retrieve information based on search terms rather than understanding the grammatical structure and context of questions.
Innovation Solution
A computer-implemented method that receives a natural language query, submits it to a lookup engine, identifies candidate sentences from search results, calculates scores based on n-gram frequency, query relevance, and document ranking, and sorts these sentences to provide a top-ranked answer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If search engines retrieve information based on search terms only, then the search process is simple and fast, but the ability to understand and answer natural language questions is poor
Solution Approach 1:
The patent segments the search query into individual sentences and processes each sentence separately through multiple scoring dimensions (n-gram frequency, document ranking, sentence length, commonality). This segmentation allows the system to handle complex natural language questions by breaking them down into manageable units while maintaining overall system effectiveness.
Solution Approach 2:
The patent introduces an intermediary scoring mechanism that bridges traditional keyword-based search and natural language understanding. By calculating multiple scores (n-gram frequency, document ranking, sentence commonality) and combining them, the system creates an intermediate layer that translates natural language questions into structured search results without requiring full AI-based language understanding.
2Measurement precision
If the search engine returns all search results, then completeness is high, but the user cannot easily identify the most relevant answer
Solution Approach 1:
The patent applies local quality by differentiating the treatment of individual sentences within search results. Instead of treating all results uniformly, it calculates specific scores for each sentence based on n-gram frequency, document ranking, and commonality with the query. This allows the system to identify and highlight the most relevant sentences while filtering out less useful information.
Solution Approach 2:
The patent changes the parameter of result presentation by transforming unstructured search results into a ranked list of scored sentences. By introducing multiple scoring parameters (n-gram frequency, document ranking, sentence length, commonality) and combining them into an overall score, the system reorganizes the quantity and quality of results to prioritize the most relevant answers.
3Measurement precision
If the system calculates multiple scoring factors for each candidate sentence, then answer accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing n-gram frequencies in a database before actual search queries are processed. This preprocessing step creates a ready-to-use reference that speeds up the scoring process during actual searches, reducing the computational burden while maintaining scoring accuracy.
Solution Approach 2:
The patent implements partial action by selectively applying different scoring factors based on the query type and available resources. Not all scoring dimensions are equally weighted or applied to every query, allowing the system to balance accuracy with processing efficiency by focusing computational effort on the most important scoring factors.
Data Source
AI summary
Embodiments relate to systems for, and methods of, answering complex questions with mined natural language search results. The systems and methods generally accept a natural language query, retrieve initial search results for the query, and then thin the retrieved results such that acceptable natural language replies to the query are identified. The natural language results may be further processed for presentation to the requestor.