Answer Generation System Using Clustering and Sentence Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current question answer (QA) systems are limited in providing answers to user queries due to variations in question wording, order, and semantic constraints, failing to address questions without exact matches in their knowledge graph.
Innovation Solution
A system that processes user questions by extracting relevant answers from a QA knowledge base and product comment data, using word embedding, clustering, and sentence similarity measurement to generate answers by filtering and selecting from comment subsets based on similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current QA systems use exact matching in knowledge graph, then answer precision is maintained, but adaptability to question variations deteriorates
Solution Approach 1:
The system transforms questions from multiple parameters (wording, order, semantics) into a unified vector space representation using word embeddings. This parameter transformation allows exact matching in the embedded space while maintaining adaptability to surface-level variations in the original question forms.
Solution Approach 2:
Word embeddings serve as an intermediary layer between the original question text and the knowledge graph. The embedding process acts as a mediator that converts diverse question formulations into a common representation that can be precisely matched against stored answers.
2Adaptability or versatility
If QA systems expand to handle question variations, then adaptability improves, but measurement precision deteriorates
Solution Approach 1:
By changing the representation parameters of questions through word embeddings, the system maintains precise measurement capabilities while handling variations. The embedding process preserves semantic information while being invariant to superficial changes in question formulation.
3Reliability
If system uses clustering and filtering processes, then answer relevance improves, but device complexity increases
Solution Approach 1:
The answer generation process is segmented into distinct stages: embedding generation, clustering, filtering, and selection. Each stage processes a specific aspect of the question-answer matching problem, making the overall complex system manageable through modular decomposition.
Solution Approach 2:
The system performs preliminary actions by pre-computing word embeddings and clustering them before actual query processing. This preliminary organization of semantic space enables faster and more relevant answer retrieval during runtime without increasing operational complexity.
Data Source
AI summary
A method and system for generating an answer to a question. The system includes a computing device. The computing device has a processor and a storage device storing computer executable code. The computer executable code, when executed at the processor, is configured to: receive the question inputted by a user regarding a product; extract target answers from a question-answer (QA) knowledge base using the question to form an answer set; extract user comments regarding the product from a comment database to form a comment set; cluster keywords in the answer set to obtain cluster centers; filter the answer set and the comment set using the cluster centers to obtain answer subset and comment subset; and generate the answer to the question from the comment subset, wherein the answer is selected from the comment subset and has high sentence similarity to sentences in the answer subset.