Query Classification Using Hierarchical Phrase Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing chatbot technologies face challenges in efficiently and accurately determining user intent from diverse and unclear user queries, leading to difficulties in providing quick and relevant responses, which can result in user inconvenience and decreased chatbot utilization.
Innovation Solution
An electronic device equipped with a processor that analyzes user queries by identifying main queries and corresponding responses, using phrase and response filtering modules to determine the relevance of phrases and responses based on similarity scores and predefined contexts, thereby providing optimized responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the type of main query in the database is increased to more clearly classify user queries, then the accuracy of query classification is improved, but the complexity of database construction and management increases
Solution Approach 1:
The patent segments the query classification process into multiple hierarchical levels. Instead of using a single flat database with numerous query types, the system divides classification into primary query categories and sub-categories, allowing for more precise classification while maintaining manageable database structures at each level.
Solution Approach 2:
The patent introduces a hierarchical dimension to the query classification database. By organizing query types in multiple levels (primary categories, sub-categories, and specific query patterns), the system achieves higher classification precision without proportionally increasing overall database complexity, as each hierarchical level handles a specific aspect of classification.
2Loss of information
If multiple main queries are provided to obtain detailed information, then the completeness of information gathering is improved, but the time required for user response increases
Solution Approach 1:
The patent applies preliminary action by pre-processing and analyzing the user's original query to identify multiple potential intents and required information types before generating follow-up questions. The system determines in advance which information gaps need to be filled and structures follow-up queries to efficiently gather the necessary details in minimal turns.
Solution Approach 2:
The patent introduces an intermediary analysis layer between the user's original query and the follow-up questions. This intermediary component analyzes the query context, identifies implicit information needs, and generates optimized follow-up questions that efficiently gather multiple pieces of information simultaneously, reducing the total number of interaction turns required.
3Adaptability or versatility
If the number of main queries matched with user query is increased, then the coverage of query types is improved, but the recognition rate is degraded due to similar meaning duplications
Solution Approach 1:
The patent applies local quality by making each query category in the hierarchical database have specific, well-defined characteristics and scope. Instead of having overlapping query types throughout the database, each category is designed with distinct boundaries and specific matching criteria, ensuring that similar queries are properly differentiated at their specific classification level while maintaining overall coverage.
4Measurement precision
If additional steps are required to provide detailed replies, then the accuracy of response is improved, but user convenience is degraded
Solution Approach 1:
The patent applies dynamics by making the interaction process adaptive and flexible. The system dynamically adjusts the number and type of follow-up questions based on the clarity and completeness of the user's original query. When the initial query is clear and complete, the system provides immediate accurate responses without requiring additional steps. When information is missing or ambiguous, the system selectively introduces follow-up questions only for the specific information gaps identified, maintaining user convenience while ensuring response accuracy.
Data Source
AI summary
A method of controlling an electronic device includes, based on obtaining first text information corresponding to a user query, identifying a main query corresponding to the first text information; obtaining a plurality of responses related to the main query; identifying a plurality of phrases included in the first text information; identifying at least one first phrase corresponding to the main query among the plurality of phrases based on a similarity between the plurality of phrases and the main query; identifying, among the plurality of responses, at least one first response corresponding to a remaining second phrase, among the plurality of phrases except the at least one first phrase, based on a similarity between each of the plurality of responses and the remaining second phrase; and providing second text information corresponding to a remaining second response other than the at least one first response among the plurality of responses.