ChatGPT inconsistently identifies the desired class in a natural language question, often selecting the supper class or subclass.

[candlecao]

This follows the #29.
This issue is a detailed offshoot of that issue.
This issue is against the backdrop of "Subgraph Extraction from Ontology".

For example, the NLQ is 广西壮族自治区(作为地域类)的音乐类型(乐种)会使用哪些乐器，那些乐器又分别对应哪些民族？
(translated as: Which musical instruments are used in the music types of Guangxi Zhuang Autonomous Region (GZAR) (as a Place), and which ethnic groups do those instruments correspond to?)

LLMs extracted places:Province instead of its supper class bf:Place which is expected for subgraph extraction.
Refer to:

[candlecao]

Solution 1: Manually adjust the ontology by adding child-class to the rdfs:range of the associated object property. For the upper case, the associated property is bf:place:

comment: (1)It's more of a temporary expedient (2) This adjustment doesn't cause collide with the original semantic meaning, however, (3) it's advisable to matain two versions of ontology: one is like this, tailored for subgraph extraction from OWL; the other one is the intact ontology for schema description and knowledge reasoning