AI 용어집
인공지능 완전 사전
Structured Knowledge Base
A set of data organized according to a predefined schema (e.g., knowledge graphs, relational databases) used to store and retrieve factual information for the QA system.
Named Entity Recognition (NER)
An NLP process that identifies and categorizes key information (names of people, organizations, drugs, etc.) in unstructured text to populate the knowledge base.
Synthetic Question Generation
A data augmentation technique where questions and their answers are automatically created from existing documents to train QA models with limited real data.
Knowledge Graph Reasoning
The ability of a QA system to infer new information by traversing and combining the relationships (edges) between entities (nodes) in a knowledge graph.
Query Rewriting
The process of transforming a user's question, often ambiguous or poorly phrased, into an optimized query for information retrieval or querying a knowledge base.
Fact Checking
A mechanism that validates the consistency and accuracy of a generated answer by cross-referencing it with reliable domain sources, crucial for high-risk applications like medicine or finance.
Domain-Specific Language Model (Domain-Specific LM)
A language model pre-trained or fine-tuned on a corpus of texts specific to a domain (e.g., medical articles, case law), enabling it to understand and generate technical jargon with greater precision.
Hybrid QA System
An architecture that combines multiple approaches (e.g., rule-based extraction, semantic search, neural models) to leverage their respective strengths and improve the robustness and accuracy of answers.
Answer Explainability
The ability of a QA system to provide not only an answer but also the evidence, sources, or reasoning that led to that answer, essential for user trust.
Semantic Indexing
Technique that organizes documents based on their meaning and context (often via embeddings), rather than on keywords, to enable more relevant search.
Information Processing Pipeline
Ordered sequence of steps (e.g., ingestion, cleaning, NER, indexing) that transforms raw data into a structured and queryable knowledge base for the QA system.
Fine-tuning on QA Data
Process of adapting a pre-trained language model using a dataset of domain-specific question-answer pairs to improve its performance on the question-answering task.
Question Intent Detection
Classification of the type of information sought by the user (e.g., definition, comparison, list, cause-effect) to guide the search and response generation strategy for the most appropriate answer.