Multi-hop Reasoning
A reasoning approach that derives answers by chaining multiple inference steps or traversing multiple knowledge nodes in sequence. Commonly required in complex question-answering tasks, multi-hop reasoning presents a challenge to traditional single-step retrieval methods.
Core Facts
Timeline (last 90 days)
In multi-hop reasoning tasks, the success rate of retrieval-type tasks (single-hop) is far higher than that of reasoning-type tasks (multi-hop).
The answers to multi-hop reasoning queries cannot be derived directly from a single passage but require chained inference across two or more document segments, making standard single-round retrieval largely unsuccessful for such queries.
In multi-hop questions, answers are scattered across multiple documents and require finding clue A before knowing to look for clue B; the agent loop transforms retrieval from a one-time action into a repeatedly executable tool call.
Multi-hop Reasoning refers to answering a question that requires traversing multiple independent documents or knowledge nodes, using intermediate reasoning results as input for the next retrieval step.
In multi-hop reasoning scenarios, each hop's retrieval should incrementally supplement based on the conclusions of the previous hop, and the decision model can explicitly track the known information set to avoid redundant recall.
Pure vector RAG is prone to a dilemma in scenarios requiring multi-hop reasoning or precise attribution, where relevant content is retrieved but cannot be explained.
Vector search was not designed from the outset to represent relationships and multi-hop reasoning.
The accuracy of multi-hop reasoning is highly dependent on the reliability of low-hop (especially single-hop) factual foundations.
This arXiv paper points out that if large language models are trained on isolated triples, they tend not to learn the surrounding context of those triples, which is necessary for multi-hop reasoning.
Multi-hop reasoning refers to complex questions that require sequentially or jointly extracting information from multiple sources to answer; implementation methods include iterative retrieval, graph-structured retrieval, and agent-based approaches, which improve complex question answering but increase latency and the probability of errors.
3 more timeline events
All Facts (13)
Knowledge graphs use a graph-based structure consisting of nodes representing entities and edges representing relationships.
65%UnverifiedIn multi-hop reasoning tasks, the success rate of retrieval-type tasks (single-hop) is far higher than that of reasoning-type tasks (multi-hop).
50%UnverifiedThe answers to multi-hop reasoning queries cannot be derived directly from a single passage but require chained inference across two or more document segments, making standard single-round retrieval largely unsuccessful for such queries.
50%UnverifiedIn multi-hop questions, answers are scattered across multiple documents and require finding clue A before knowing to look for clue B; the agent loop transforms retrieval from a one-time action into a repeatedly executable tool call.
50%UnverifiedMulti-hop Reasoning refers to answering a question that requires traversing multiple independent documents or knowledge nodes, using intermediate reasoning results as input for the next retrieval step.
50%UnverifiedIn multi-hop reasoning scenarios, each hop's retrieval should incrementally supplement based on the conclusions of the previous hop, and the decision model can explicitly track the known information set to avoid redundant recall.
50%UnverifiedPure vector RAG is prone to a dilemma in scenarios requiring multi-hop reasoning or precise attribution, where relevant content is retrieved but cannot be explained.
50%UnverifiedVector search was not designed from the outset to represent relationships and multi-hop reasoning.
50%UnverifiedThis arXiv paper points out that if large language models are trained on isolated triples, they tend not to learn the surrounding context of those triples, which is necessary for multi-hop reasoning.
50%UnverifiedThe accuracy of multi-hop reasoning is highly dependent on the reliability of low-hop (especially single-hop) factual foundations.
50%UnverifiedMulti-hop reasoning refers to complex questions that require sequentially or jointly extracting information from multiple sources to answer; implementation methods include iterative retrieval, graph-structured retrieval, and agent-based approaches, which improve complex question answering but increase latency and the probability of errors.
50%UnverifiedPerforming correlation analysis across multiple documents involves multi-hop reasoning, which is a known weak point of current RAG architectures.
50%Unverified多跳推理指从起始节点出发经过两个或以上关系边到达目标节点的推理过程，在图数据库中通过图遍历算法（如BFS、Dijkstra）或图查询语言（如Cypher、SPARQL）实现
50%