AI-ordlista
Den kompletta ordlistan över AI
Multi-hop Question-Answering
Subfield of question-answering systems where answering an initial question requires formulating and solving a sequence of intermediate questions, often chained, to arrive at the final answer.
Chain Reasoning
Cognitive and computational process that involves linking multiple facts or entities to infer new information, fundamental for multi-hop QA systems that must follow logical paths in a knowledge base.
Question Decomposition
Technique involving analyzing a complex question to split it into a series of simpler and directly answerable sub-questions, whose answers will then be combined to solve the initial question.
Retrieval-Augmented Generation (RAG)
Hybrid architecture where a retriever model finds relevant documents to feed a generative model, often used to build complex responses by relying on external sources.
Query Planning
Mechanism that determines the optimal order of sub-queries to execute to answer a multi-hop question, minimizing computational costs and maximizing the relevance of retrieved information.
Evidence Fusion
Final step of the multi-hop QA process where information from different reasoning stages is synthesized and consolidated to formulate a coherent and complete final answer.
Ambiguous Entity
Major challenge in multi-hop QA where an entity mentioned in the question may refer to multiple distinct nodes in the knowledge graph, requiring contextual disambiguation to follow the correct reasoning path.
Path Reasoning
Method specific to multi-hop QA that involves exploring and evaluating different relational paths in a knowledge graph to find the sequence of logical hops leading to the answer.
State Transition Model
A formal approach to modeling the multi-hop reasoning process as a series of transitions between states (known information), where each action (reasoning hop) modifies the system's current state.
Reinforcement Learning for QA
A training paradigm where a QA agent learns a navigation policy (which questions to ask or which links to follow) by being rewarded when it reaches the correct answer, thereby optimizing its reasoning strategy.
Symbolic Neural Question Answering
A hybrid approach combining neural networks for natural language understanding and symbolic components for logical reasoning and manipulation of structured facts, particularly well-suited for multi-hop QA.
Reasoning Explainability
The ability of a multi-hop QA system to not only provide an answer but also to expose the chain of deductions, sources, and intermediate steps that led to that answer, essential for trust and debugging.
Relation Prediction
The task of identifying the correct semantic relation between two entities in a knowledge graph, a key step in each 'hop' of a multi-hop question-answering process.
Document-Based QA System
A type of multi-hop QA system where reasoning is performed not on a structured graph, but by connecting information scattered across a collection of unstructured or semi-structured documents.