Semantic QA Retrieval With LLM Prompting for Broader Question Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering (QA) systems face limitations in accuracy, coverage, and scalability due to their reliance on precise technical terminology, inability to understand nuanced human language, and the need for extensive human intervention to update knowledge sources, leading to inaccurate and inconsistent answers.
Innovation Solution
A system leveraging Large Language Models (LLMs) and multi-agent Retrieval-Augmented Generation (RAG) to integrate explicit, relevant information from external knowledge sources, enhancing accuracy and contextually rich answers by combining LLMs with retrieval of specific data, and using hybridized retrieval and prompt optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If keyword-based search systems are used to answer questions, then the system can operate automatically without human input, but the system suffers from limited coverage and can only answer questions where pre-existing exact match question and answer pairs are available
Solution Approach 1:
The patent replaces traditional keyword-based mechanical search systems with a semantic embedding-based neural network system. The system uses sentence transformers to convert questions and search queries into vector representations, enabling semantic similarity search that understands meaning rather than relying on exact keyword matches. This substitution resolves the contradiction by maintaining automated operation while dramatically improving question coverage and adaptability.
Solution Approach 2:
The patent changes the fundamental parameter of search from exact keyword matching to semantic similarity measurement using cosine similarity on vector embeddings. This parameter change allows the system to answer questions that are semantically equivalent even if they use different wording, thereby improving coverage while maintaining automation.
2Measurement precision
If users must search using precise technical terminology to get accurate results, then the system can provide accurate answers, but the system becomes inaccessible to average members of society who do not know the correct technical terms
Solution Approach 1:
The patent replaces precise keyword matching with semantic similarity search using neural network embeddings. This allows users to ask questions in natural language without knowing technical terminology, while the system automatically finds semantically equivalent questions in the database. The substitution maintains answer accuracy while dramatically improving ease of operation and user accessibility.
Solution Approach 2:
The patent introduces sentence transformers and vector embeddings as an intermediary between user queries and the question database. This intermediary layer translates natural language queries into semantic representations that can be matched against stored questions, bridging the gap between user intent and precise information retrieval without requiring users to know technical terms.
3Measurement precision
If deep learning models are used to improve accuracy by leveraging patterns from large datasets, then the system can handle complexity and nuance of human language, but the systems still struggle with the breadth of human language and context-dependent meanings
Solution Approach 1:
The patent uses a universal sentence transformer model that can process any natural language question regardless of domain or complexity. The embedding model is trained on diverse datasets including SQuAD, making it universally applicable to different types of questions while maintaining high accuracy. This universality resolves the contradiction by enabling both precise language understanding and broad language coverage.
Solution Approach 2:
The patent performs preliminary action by pre-training the sentence transformer on large diverse datasets (including 25 GB of SQuAD data) before deployment. This pre-training equips the model with broad language understanding capabilities that enable it to handle various types of questions accurately without requiring domain-specific fine-tuning for each new application.
4Reliability
If IBM Watson and similar systems rely heavily on custom programming for grammar and syntax analysis along with large hand-curated knowledge bases, then the systems can provide useful answers, but the approach does not generalize well and requires large and fast processing capabilities
Solution Approach 1:
The patent replaces complex custom programming for grammar and syntax analysis with a pre-trained neural network model that automatically understands language structure. The sentence transformer handles all linguistic analysis internally through its training, eliminating the need for separate custom programming layers. This substitution maintains answer reliability while significantly reducing system complexity and removing the need for large hand-curated knowledge bases.
Solution Approach 2:
The patent uses copying by leveraging pre-trained models and architectures developed by other researchers (such as sentence transformers from Hugging Face). Instead of building language understanding capabilities from scratch, the system copies and adapts proven neural network architectures, reducing development complexity while maintaining high reliability.
Data Source
AI summary
An information retrieval and summarization system has a knowledge database that stores a plurality of data records, each having a data document ID, data document title, data source link, and data document text retrieved from the data source link. A question and answer (QA) database stores a QA ID, QA title, and an analyst-generated question and answer (QA) pair. The analyst-generated QA pair has an analyst-generated question and an analyst-generated answer associated with the analyst-generated question. A custom prompt database stores a prompt record that includes a prompt instruction, prompt clarification, and a predefined prompt example. A QA generation processing device applies the prompt record to the data document text stored in said knowledge database to generate a plurality of automated-generated QA pairs using a large language model, and store the plurality of automated-generated QA pairs in the QA database.