A DeBERTa-v3-large model fine-tuned for Relation Extraction (RE) on scholarly documents that mention software. This model identifies semantic relationships (e.g., Developer_of, Version_of) between software-related entities in academic text.

🧪 Training Data

This model was trained on the following dataset:

psresearch/NER-RE-for-Software-Mentions

wanted to load and run this model check this - submission_recreate.ipynb The dataset contains annotated relationships between named entities found in scholarly papers related to software engineering.

📊 Metrics

on testset

Relation	Precision	Recall	F1-Score	Support
Developer_of	0.2344	0.7500	0.3571	20
Citation_of	0.5321	0.7968	0.6381	187
Version_of	0.3901	0.7396	0.5108	96
PlugIn_of	0.1013	0.6154	0.1739	13
URL_of	0.4701	0.7857	0.5882	70
License_of	0.0000	0.0000	0.0000	0
AlternativeName_of	0.6522	0.8824	0.7500	17
Release_of	0.5263	1.0000	0.6897	10
Abbreviation_of	0.5000	0.5000	0.5000	12
Extension_of	0.0000	0.0000	0.0000	6
Specification_of	0.0000	0.0000	0.0000	0
Micro Avg	0.4240	0.7633	0.5452	431
Macro Avg	0.3785	0.6744	0.4675	431
Weighted Avg	0.4599	0.7633	0.5675	431

📈 Model Comparison

Task	Model / Setup	Precision	Recall	F1
RE	DeBERTa-V3-Large	0.1025	0.4117	0.1543
RE	Modern BERT-Large	0.0878	0.4228	0.1379
RE	DeBERTa-V3-Large (Augmented Data)	0.3785	0.6744	0.4675
RE	Modern BERT-Large (Augmented Data)	0.3473	0.6702	0.4384

🏷️ Label Mapping

python

{
  "Developer_of": 0,
  "URL_of": 1,
  "Version_of": 2,
  "Citation_of": 3,
  "PlugIn_of": 4,
  "Extension_of": 5,
  "Specification_of": 6,
  "no_relation": 7,
  "Release_of": 8,
  "Abbreviation_of": 9,
  "License_of": 10,
  "AlternativeName_of": 11
}