0.25
0.5
0.75
1.25
1.5
1.75
2
Kastor: Fine-tuned Small Language Models for Shape-based Active Relation Extraction
Related categories
Presentation
Kastor - Fine-tuned Small Language Models for Shape-based Active Relation Extraction00:00
Research questions:00:15
Kastor00:43
RDF-Pattern Based Extraction01:06
RDF-Pattern Based Extraction01:25
RDF-Pattern Based Extraction01:53
Example-specific patterns of a maximal shape02:13
Example-specific patterns of a maximal shape02:42
Example-specific patterns of a maximal shape03:00
Example-specific patterns of a maximal shape03:03
Example-specific patterns of the maximal shape dbo:Person03:10
RDF-Pattern Based Extraction03:25
Kastor03:48
Knowledge distillation04:07
Rule augmentation04:28
Knowledge distillation04:43
Wikicheck04:51
Knowledge distillation05:16
Distilled KG Example-specific pattern caracterisation05:36
Distilled KG Example-specific pattern caracterisation05:57
Kastor06:09
Finetuning details06:21
Light Active learning06:42
Light Active learning07:14
Light Active learning07:44
Light Active learning07:52
Light Active learning07:56
Light Active learning08:10
Light Active learning08:21
Light Active learning08:45
Kastor08:55
Model evaluation09:00
Model evaluation09:27
Model evaluation09:45
Model evaluation09:55
Model evaluation10:02
Annotation results10:22
Annotation results10:39
Annotation results10:51
Annotation results11:04
Errors (FP-) classification11:23
Errors (FP-) classification11:30
Errors (FP-) classification11:38
Errors (FP-) classification11:45
Errors (FP-) classification11:53
Errors (FP-) classification12:19
Results12:29
So how to avoid FP- ?12:39
Results12:56
Kastor13:13
CONCLUSIONS13:19
An open framework for systematic and auditable RE with KG and SLM13:33
A lot of other details could be find in the papers :13:55