KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Conversational RAG
Hybrid architecture combining the retrieval of relevant information from a knowledge base with the generation of contextual responses by an LLM, specifically adapted for dialog interactions.
Conversational Context Window
Buffer memory containing the recent dialogue history, used to maintain consistency in generated responses and to refine the retrieval query in a RAG system.
Query Rewriting
Process that transforms an implicit or contextual user question into an explicit and optimized query for vector search in the knowledge base.
Conversational Reranking
Post-processing step that reorders retrieved documents based on their relevance not only semantically but also contextually relative to the dialogue history.
Context Fusion
Mechanism for intelligently integrating retrieved information with conversational history to build a coherent and complete prompt for the generation model.
Retrieval-Augmented Generation
Response generation process where the LLM uses both its intrinsic knowledge and the specific retrieved documents to produce an accurate and contextualized response.
Conversational Hallucination Detection
Validation system that checks whether the generated response is well-supported by the retrieved documents, crucial for maintaining reliability in RAG dialogues.
Conversational Metadata
Structured information about the dialogue (intents, entities, sentiment) used to guide retrieval and generation, improving response relevance in a RAG context.
Strategic Chunking
Intelligent segmentation of documents into optimized segments for retrieval, taking into account semantic structure and specific needs of conversational interactions.
Conversational RAG Prompt
Specialized prompt structure that integrates instruction, dialogue history, retrieved documents, and formatting constraints to guide response generation.
Multi-level Indexing
Indexing architecture that organizes documents according to multiple granularities (paragraphs, sections, documents) to enable flexible retrieval based on dialogue needs.
Conversational Feedback Loop
Continuous learning mechanism that uses user interactions to refine retrieval and generation strategies, progressively improving RAG system performance.
Dynamic Citation
System capability to automatically reference specific sources used to generate each part of the response, essential for transparency and verification in conversational RAG.
Contextual Ambiguity Management
Process that identifies and resolves ambiguous references in dialogue using conversational history to clarify intentions before information retrieval.
Conversational Vector Store
Vector database optimized for conversational workloads, with features like metadata filtering and hybrid text-vector search.