33、Text2KGBench：文本到知识图谱生成的基准测试

Text2KGBench：文本到知识图谱生成的基准测试

在自然语言处理领域，从文本中生成知识图谱是一项重要的任务。本文将介绍Text2KGBench这一基准测试，它用于评估大语言模型（LLM）在本体引导下从文本语料中提取事实的能力。

1. 数据集统计

首先来看两个数据集的统计信息，如下表所示：
| wikidata - tekgen | | | | dbpedia - webnlg | | | |
| — | — | — | — | — | — | — | — |
| Ontology | Types | Rels. | Sents. | Ontology | Types | Rels | Sents. |
| Movie | 12 | 15 | 2800 | University | 15 | 46 | 156 |
| Music | 13 | 13 | 2243 | Music | 15 | 35 | 290 |
| Sport | 15 | 11 | 1693 | Airport | 14 | 39 | 306 |
| Book | 20 | 12 | 1810 | Building | 14 | 38 | 275 |
| Military | 13 | 9 | 750 | Athlete | 17 | 37 | 293 |
| Computer | 15 | 12 | 743 | Politician | 19 | 40 | 319 |
| Space | 15 | 7 | 666 | City | 11 | 23 |