开发基于知识的问答系统时，何时需使用Jena这类推理机？

Great question—this is a super common consideration when building QA systems that rely on knowledge graphs. From my experience, Jena (or similar rule-based/OWL reasoners) are most valuable in two core scenarios that align with your thinking:

• Fallback reasoning for online query gaps
  Start by querying your raw RDF triples directly with SPARQL—this is fast and straightforward for explicit facts. But when that query returns no results, fire up Jena to derive implicit answers. For example: if your graph only has :Alice :parentOf :Bob and the rule :parentOf owl:inverseOf :childOf, a direct query for :Bob :childOf ?who won’t hit any explicit triples. Jena can infer that missing relationship on the fly, letting your QA system return a correct answer instead of a "no result" message.

• Offline knowledge graph completion to optimize online performance
  Instead of running inference every time a user asks a question, use Jena offline to precompute all possible implicit triples and add them to your knowledge graph. This is a huge win for high-traffic QA systems where low latency matters. For instance, if you have a class hierarchy like :Dog rdfs:subClassOf :Mammal, Jena can pre-infer that every :Dog instance is also a :Mammal instance. When users ask "What are all mammals?", your system just runs a simple SPARQL query against the pre-completed graph instead of triggering real-time inference, cutting down response time significantly.

内容的提问来源于stack exchange，提问作者pun