Godfrey2712/amf_illoc_force_intent_recognition

This model can be used to classify communicative intentions in Dialogue. It takes a single sentence/utterance as input and returns an illocutionary force label.

Below are the labels used to fine-tune a RoBERTa-large model:

"0": "Agreeing",  
"1": "Arguing",  
"2": "Asserting",  
"3": "Assertive Questioning",  
"4": "Challenging",  
"5": "Default Illocuting",  
"6": "Disagreeing",  
"7": "Pure Questioning",  
"8": "Restating",  
"9": "Rhetorical Questioning"  

F1_macro - 0.55

F1_weighted - 0.67

Accuracy - 0.67

These are the links to the Datasets used:

MM2012 - https://corpora.aifdb.org/mm2012

MM123 - https://corpora.aifdb.org/mm123

QT30 - https://corpora.aifdb.org/qt30

US2016 - https://corpora.aifdb.org/US2016

	precision	recall	f1-score	support
Agreeing	0.63	0.67	0.65	18
Arguing	0.67	0.40	0.50	5
Asserting	0.79	0.89	0.84	200
Assertive Questioning	0.52	0.44	0.48	124
Challenging	0.48	0.42	0.44	24
Default Illocuting	1.00	0.07	0.13	14
Disagreeing	0.33	0.40	0.36	5
Pure Questioning	0.75	0.76	0.76	293
Restating	1.00	1.00	1.00	2
Rhetorical Questioning	0.36	0.40	0.38	81

accuracy			0.67	766
macro avg	0.65	0.54	0.55	766
weighted avg	0.67	0.67	0.67	766

The data preprocessing and fine-tuning technique can be found here: https://discovery.dundee.ac.uk/en/studentTheses/exploiting-illocutionary-forces-in-dialogue-structures-for-enhanc/

Citation:

Inyama, G. (2025). Exploiting Illocutionary Forces in Dialogue Structures for Enhancing Authorship Identification [Master of Philosophy thesis, University of Dundee]. Discovery - the University of Dundee Research Portal. https://doi.org/10.15132/20000713