Query Parsing for Search Accuracy in Ambiguous Sentences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search technologies face challenges in providing accurate search results for complex query sentences due to high ambiguity and difficulty in considering logical relationships between categories and meanings within the query.
Innovation Solution
An electronic device and method that parse query sentences to extract semantic phrases, conditional phrases, and relationship information, converting them into embedding vectors to compare with search target data, and determine search results based on similarity and metadata comparisons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If embedding vector-based neural network search technology is used, then search capability is provided for various query sentences, but search accuracy deteriorates when query sentences have high ambiguity with multiple meanings
Solution Approach 1:
The query sentence is segmented into multiple components including semantic phrases, conditional phrases, and relationship information. Each component is processed separately through parsing to extract meaningful elements, allowing the system to handle ambiguous queries by analyzing their structural composition rather than treating them as monolithic inputs
Solution Approach 2:
Embedding vectors serve as intermediary representations that bridge the gap between parsed query components and search target data. The system converts semantic phrases into embedding vectors and compares them with indexed vectors to find similar items, using these vectors as mediators to resolve ambiguity in the original query language
2Device complexity
If simple embedding vector comparison is used, then search process is simplified, but the ability to consider logical relationships between query categories and meanings is lost
Solution Approach 1:
The search process is segmented into distinct phases: parsing the query into semantic and conditional components, converting to embedding vectors, comparing with indexed data, and finally synthesizing results based on relationship information. This segmentation allows complex logical reasoning to be distributed across multiple simple, manageable steps
Solution Approach 2:
The system performs preliminary parsing and structural analysis of the query sentence before conducting the actual search. By pre-processing the query to extract semantic phrases, conditional phrases, and relationship information, the system prepares the logical framework in advance, enabling more reliable reasoning during the search execution phase
Data Source
AI summary
An electronic device and a method for providing a search result related to a query sentence are provided. The method includes obtaining, by parsing the query sentence, at least one semantic phrase, at least one conditional phrase, and relationship information between at least two of the at least one semantic phrase and the at least one conditional phrase, converting the at least one semantic phrase into at least one first embedding vector, obtaining, as a first search result, search target data corresponding to the second embedding vector similar to the first embedding vector, obtaining, as a second search result, search target data including metadata corresponding to the at least one conditional phrase, and determines, based on the relationship information, a search result that is to be provided to the user from the first search result and the second search result.