ACL 2023

RoBERTa-Based Multi-class Emotion detection on highly imbalanced data

Detection of Emotions (Multiple) from user essays as a reaction to news articles. Existing works focus on single class cases , while multiple emotions may be exibited in a single text.

Finished 1st out of 84 teams : WASSA 2023 Task 3

  • Usually multi-way single class classifications or text boundary predictions between different emotions larger texts

  • Working on Highly Imbalanced datasets. Including cases where multiple emotions are conveyed in one text

  • Adding Multilingual aspect , detecting which part of the text convey what emotion

NAACL 2024

Black-Box Word-Level Text Boundary
Detection in Partially Machine Generated Texts

Word level text boundary detection in texts where a portion is machine generated and rest is human written. Proposed models performed better than existing proprietary systems. Well suited for text-completion and instruct models. Works on unseen domains and generators too

Finished 1st out of 308 teams : SemEval 2024 Task 8C

  • Current proprietary systems classify a text as entirely-human written or entirely-machine generate. Since used data domains are all academic , it is assumed the task is aimed at academic usage. Obviously people are not retarted to just copy paste text from ChatGPT. Some of the existing works display what parts are likely machine generated with a very low accuracy sometimes below 50%

  • Same performance irrespective of text length while existing proprietary systems struggle with shorter texts. Almost 25% higher accuracy than existing proprietary systems.

  • Multilingual extension , applying to social media texts

LREC-COLING 2024

Can The Potential for Offline Harm Events in social media texts Be Identified Without the Context?

Comparison of various approaches and methods and their effectiveness in detecting target groups and possibility of texts leading to riots/clashes/protests

Finished 3rd out of ?? teams : HarmPot 2024 Task 1

  • Description text goes here

  • Description text goes here

  • Description text goes here

ACL 2024

Placeholder

placeholder

  • Description text goes here

  • Description text goes here

  • Description text goes here

ACL 2024

Placeholder

Description

  • Description text goes here

  • Description text goes here

  • Description text goes here