里放背景部分:Graph-based retrieval-augmented generation depends on the quality of the knowledge graph (KG) it retrieves over, yet schema-free LLM-based extraction pipelines typically produce flat graphs: deduplicated entities and relations with no type hierarchy and no domain/range constraints.
然后里放介绍AutoOntic的:We present AutoOntic, a pipeline that automatically constructs an ontology from text by canonicalizing open-vocabulary entity types and relations, inducing a hierarchy over the canonicalized types, and mining relation-level domain and range constraints.
然后下一个讲它的层次归纳过程:Its hierarchy induction procedure prioritizes types using corpus-derived relation-argument profiles and places them through local, LLM-adjudicated parent/child checks, allowing newly inserted types to trigger reassignment of existing parent-child links.
然后下一个讲结果的作用:The resulting hierarchy constrains entity deduplication and supports the induction of relation-level domain/range constraints.
然后下一个讲评估方法:For intrinsic evaluation, we introduce a reference-based protocol that classifies induced ontology types as exact, broader, narrower, or unrelated to reference domain and range classes.
然后下一个讲第一个实验结果:On the Wikidata-TekGen and DBpedia-WebNLG subsets of Text2KGBench, over 90% of AutoOntic's induced endpoint types are semantically aligned with the reference classes across two LLM backbones: about half match exactly, while most remaining differences concern granularity.
然后下一个讲第二个实验结果:On MINE, knowledge graphs organized using an ontology built by AutoOntic achieve 88% retrieval accuracy.
然后最后一个讲可用地址:AutoOntic is available at https://github.com/screemix/AutoOntic.
不对,但是这样会不会改变了原文的段落?哦,原文的纯文本没有分段,那是不是应该保持成一个p?不对,等下看规则2:每个自然段落用包裹,长文用h2/h3划分小节。原文是学术类的摘要,没有小节,所以不用h2/h3。那如果原文是一个自然段落,就用一个p?但这样排版不太语义化?不对,再仔细看原文的逻辑,其实是连贯的,但可以按语义拆分段落吗?不对,规则说不得增删改内容,但是段落划分是排版的一部分,不算改内容。哦,对,排版的时候可以把逻辑上不同的部分分成不同的p,只要内容不变。 等下,再检查有没有需要用其他标签的,比如没有列表、引用、代码,所以不用ul、blockquote、pre。然后重要的句子或者关键词用strong,比如AutoOntic是重要的,还有关键数据over 90%、88%可以加strong。 现在整理一下,确保所有内容都在,没有增删: 首先,第一个p放背景:
Graph-based retrieval-augmented generation depends on the quality of the knowledge graph (KG) it retrieves over, yet schema-free LLM-based extraction pipelines typically produce flat graphs: deduplicated entities and relations with no type hierarchy and no domain/range constraints.
然后第二个p介绍AutoOntic:We present AutoOntic, a pipeline that automatically constructs an ontology from text by canonicalizing open-vocabulary entity types and relations, inducing a hierarchy over the canonicalized types, and mining relation-level domain and range constraints.
第三个p讲层次归纳过程:Its hierarchy induction procedure prioritizes types using corpus-derived relation-argument profiles and places them through local, LLM-adjudicated parent/child checks, allowing newly inserted types to trigger reassignment of existing parent-child links.
第四个p讲层次的作用:The resulting hierarchy constrains entity deduplication and supports the induction of relation-level domain/range constraints.
第五个p讲评估方法:For intrinsic evaluation, we introduce a reference-based protocol that classifies induced ontology types as exact, broader, narrower, or unrelated to reference domain and range classes.
第六个p讲第一个实验结果:On the Wikidata-TekGen and DBpedia-WebNLG subsets of Text2KGBench, over 90% of AutoOntic's induced endpoint types are semantically aligned with the reference classes across two LLM backbones: about half match exactly, while most remaining differences concern granularity.
第七个p讲第二个实验结果:On MINE, knowledge graphs organized using an ontology built by AutoOntic achieve 88% retrieval accuracy.
第八个p讲可用性:AutoOntic is available at https://github.com/screemix/AutoOntic.
等下,但是有没有可能原文是一个整段,那是不是应该用一个p?不对,看规则说“每个自然段落用包裹”,这里的自然段落是指原文中的自然段落,但原文的纯文本没有换行,是一段。那是不是应该用一个p?但这样排版不太好?不对,再仔细看用户给的正文,确实是连续的一段,没有换行,那是不是所有内容放在一个p里?但可以在里面加strong。比如:
Graph-based retrieval-augmented generation depends on the quality of the knowledge graph (KG