LLM Subquery Segmentation for Multifaceted Query Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Search systems struggle to provide useful results for multifaceted and noisy queries, leading to increased user interaction time and resource usage on client devices.
Innovation Solution
Implementing a method that uses a generative model like LLM to generate candidate subqueries for multifaceted queries, selecting a subset based on evaluation metrics, and obtaining search results for these subqueries to generate a comprehensive response.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If search systems process multifaceted and noisy queries as a whole, then they can maintain simple query processing logic, but they fail to determine useful search results that address all facets of the query
Solution Approach 1:
The patent applies segmentation by dividing a multifaceted query into multiple subqueries, each addressing a specific facet. The system identifies different topics or problems within the query (e.g., wireless router setup, smart thermostat, vacuum) and generates separate subqueries for each, enabling comprehensive coverage of all query aspects while maintaining simple processing logic for each individual subquery.
2Reliability
If users manually break multifaceted queries into individual queries and submit them separately, then they can obtain useful search results for each facet, but the user interaction duration and resource usage increase significantly
Solution Approach 1:
The system applies self-service by automatically performing the query decomposition task that would otherwise require manual user intervention. The automated assistant system identifies facets, generates subqueries, and submits them sequentially without requiring the user to manually break down the query or wait for forum responses, thus obtaining comprehensive results while minimizing user interaction time and device resource usage.
3Ease of operation
If LLMs process multifaceted and noisy natural language inputs directly, then they can maintain simple input processing, but the generated responses fail to address all facets effectively
Solution Approach 1:
The patent applies segmentation by dividing the complex natural language input into multiple subqueries, each targeting a specific facet. The system processes the noisy NL input by generating separate subqueries for each identified topic or problem, then obtains search results for each subquery and combines them into a comprehensive response that addresses all facets effectively.
4Adaptability or versatility
If search systems return results for all identified facets, then they can provide comprehensive coverage, but the response complexity and information overload increase
Solution Approach 1:
The patent applies segmentation by organizing search results according to the segmented subqueries. Each subquery's results are presented in a structured manner that corresponds to the specific facet it addresses, making the overall response comprehensive yet organized. This segmentation approach allows the system to provide complete coverage while maintaining manageable response structure through clear separation of different topic areas.
Data Source
AI summary
Implementations leverage a generative model (e.g., a large language model (LLM)) to generate a plurality of candidate subqueries for multifaceted natural language (NL) based input, where each of the candidate subqueries is potentially directed to a facet or problem of the multifaceted NL based input. Those implementations further select, from the plurality of candidate subqueries and using one or more evaluation metrics, a subset of the candidate queries. Those implementations further, in response to selecting the subset of the candidate queries, obtain, for each of the candidate subqueries of the selected subset, at least one corresponding search result. Those implementations further generate a response to the NL based input based on the corresponding search results for the candidate subqueries of the subset, and cause the response to be rendered responsive to the NL based input.