Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Multi-hop Retrieval
Information retrieval process requiring multiple sequential steps to fill information gaps and assemble a complete answer from multiple sources.
Iterative Retrieval
Retrieval method where the system performs successive search cycles, using previous results to refine and improve the relevance of subsequent queries.
Chain-of-Thought Retrieval
Approach where the system follows explicit step-by-step reasoning, retrieving information at each stage to progressively build a coherent answer.
Fusion Retrieval
Strategy combining results from multiple sources or retrieval methods into a unified set, weighting each contribution according to its contextual relevance.
Recursive Retrieval
Process where each retrieval step generates new queries based on previously obtained information, creating a tree-like search structure.
Hop-wise Reasoning
System's ability to perform logical inferences at each retrieval step, using acquired knowledge to guide subsequent searches.
Knowledge Graph Traversal
Navigation through nodes and relationships of a knowledge graph to discover connected information and establish complex semantic links.
Context Accumulation
Process of progressively aggregating relevant context across multiple retrieval steps, continuously enriching the system's understanding.
Retrieval Chain
Ordered sequence of retrieval operations where each link logically depends on the previous one to build a complete final response.
Reasoning Path
Logical trajectory followed by the system through different information sources to arrive at a coherent conclusion or answer.
Evidence Synthesis
Integration and consolidation of evidence collected through multiple retrieval steps to form a unified argument or response.
Progressive Retrieval
Graduated approach where the search evolves from general to specific, progressively refining the information target at each iteration.
Hierarchical Retrieval
Retrieval structure organized in hierarchical levels, where each level provides different context to refine the subsequent search.
Cross-hop Attention
Attention mechanism allowing the system to make connections between information retrieved at different steps to maintain overall coherence.
Multi-document Reasoning
Ability to reason about information distributed across multiple documents, establishing relevant correlations and contradictions.
Retrieval Graph
Structural representation of relationships between retrieved information, visualizing reasoning paths and dependencies between sources.