Multi-Hop Reasoning Task

A Multi-Hop Reasoning Task is a reasoning task that requiring multiple reasoning steps (to reach a conclusion from given evidence).

• Context:
  • Task Input: source information, context, query
    • Optional Input: reasoning path, intermediate knowledge, inference rules
  • Task Output: final conclusion, reasoning chain, supporting evidence
  • Task Performance Measure: reasoning accuracy, path validity, logical coherence
  • ...
  • It can range from being a Simple Multi-Hop Task to being a Complex Multi-Hop Task, depending on its reasoning depth.
  • It can range from being a Linear Multi-Hop Task to being a Graph Multi-Hop Task, depending on its reasoning structure.
  • It can range from being a Factual Multi-Hop Task to being an Abstract Multi-Hop Task, depending on its reasoning type.
  • It can typically require Information Connection through logical bridges.
  • It can typically perform Step-by-Step Inference through reasoning chains.
  • It can typically support Knowledge Integration through evidence combination.
  • It can often involve Path Finding through knowledge graphs.
  • It can often utilize Intermediate Conclusions for complex reasoning.
  • It can often require Evidence Validation at each step.
  • ...
• Examples:
  • Question Answering Tasks, such as:
    • Multi-Document QA Tasks, such as cross-document inference, multi-source integration, evidence aggregation, and fact verification.
    • Complex Query Tasks, such as multi-step question, bridge question, compositional query, and chain reasoning.
    • Knowledge Graph Tasks, such as path reasoning, relation inference, entity connection, and graph traversal.
  • Logic Reasoning Tasks, such as:
    • Deductive Tasks, such as logical inference, premise combination, conclusion derivation, and proof construction.
    • Inductive Tasks, such as pattern recognition, generalization formation, rule learning, and hypothesis generation.
    • Abductive Tasks, such as explanation generation, cause inference, hypothesis formation, and scenario construction.
  • Domain Reasoning Tasks, such as:
    • Scientific Reasoning Tasks, such as hypothesis testing, evidence chain, causal inference, and mechanism explanation.
    • Legal Reasoning Tasks, such as case connection, precedent chain, argument construction, and legal inference.
    • Medical Reasoning Tasks, such as diagnosis chain, treatment planning, symptom connection, and cause analysis.
  • ...
• Counter-Examples:
  • Single-Hop Reasoning Tasks, which require only one inference step.
  • Direct Answer Tasks, which don't require intermediate reasoning.
  • Simple Lookup Tasks, which lack multi-step inference.
• See: Reasoning Chain, Knowledge Graph, Inference Task, Logic Task, Evidence Integration.

---

---