@inproceedings{prama-etal-2025-computational,
title = "Computational Story Lab at {BLP}-2025 Task 1: {H}ate{S}ense: A Multi-Task Learning Framework for Comprehensive Hate Speech Identification using {LLM}s",
author = "Prama, Tabia Tanzin and
Danforth, Christopher M. and
Dodds, Peter",
editor = "Alam, Firoj and
Kar, Sudipta and
Chowdhury, Shammur Absar and
Hassan, Naeemul and
Prince, Enamul Hoque and
Tasnim, Mohiuddin and
Rony, Md Rashad Al Hasan and
Rahman, Md Tahmid Rahman",
booktitle = "Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025)",
month = dec,
year = "2025",
address = "Mumbai, India",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.banglalp-1.37/",
pages = "430--442",
ISBN = "979-8-89176-314-2",
abstract = "This paper describes HateSense, our multi-task learning framework for the BLP 2025 shared task 1 on Bangla hate speech identification. The task requires not only detecting hate speech but also classifying its type, target, and severity. HateSense integrates binary and multi-label classifiers using both encoder- and decoder-based large language models (LLMs). We experimented with pre-trained encoder models (Bert based models), and decoder models like GPT-4.0, LLaMA 3.1 8B, and Gemma-2 9B. To address challenges such as class imbalance and the linguistic complexity of Bangla, we employed techniques like focal loss and odds ratio preference optimization (ORPO). Experimental results demonstrated that the pre-trained encoders (BanglaBert) achieved state-of-the-art performance. Among different prompting strategies, chain-of-thought (CoT) combined with few-shot prompting proved most effective. Following the HateSense framework, our system attained competitive micro-F1 scores: 0.741 (Task 1A), 0.724 (Task 1B), and 0.7233 (Task 1C). These findings affirm the effectiveness of transformer-based architectures for Bangla hate speech detection and suggest promising avenues for multi-task learning in low-resource languages."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="prama-etal-2025-computational">
<titleInfo>
<title>Computational Story Lab at BLP-2025 Task 1: HateSense: A Multi-Task Learning Framework for Comprehensive Hate Speech Identification using LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tabia</namePart>
<namePart type="given">Tanzin</namePart>
<namePart type="family">Prama</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christopher</namePart>
<namePart type="given">M</namePart>
<namePart type="family">Danforth</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peter</namePart>
<namePart type="family">Dodds</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Firoj</namePart>
<namePart type="family">Alam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sudipta</namePart>
<namePart type="family">Kar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shammur</namePart>
<namePart type="given">Absar</namePart>
<namePart type="family">Chowdhury</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naeemul</namePart>
<namePart type="family">Hassan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Enamul</namePart>
<namePart type="given">Hoque</namePart>
<namePart type="family">Prince</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohiuddin</namePart>
<namePart type="family">Tasnim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Rashad</namePart>
<namePart type="given">Al</namePart>
<namePart type="given">Hasan</namePart>
<namePart type="family">Rony</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Tahmid</namePart>
<namePart type="given">Rahman</namePart>
<namePart type="family">Rahman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Mumbai, India</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-314-2</identifier>
</relatedItem>
<abstract>This paper describes HateSense, our multi-task learning framework for the BLP 2025 shared task 1 on Bangla hate speech identification. The task requires not only detecting hate speech but also classifying its type, target, and severity. HateSense integrates binary and multi-label classifiers using both encoder- and decoder-based large language models (LLMs). We experimented with pre-trained encoder models (Bert based models), and decoder models like GPT-4.0, LLaMA 3.1 8B, and Gemma-2 9B. To address challenges such as class imbalance and the linguistic complexity of Bangla, we employed techniques like focal loss and odds ratio preference optimization (ORPO). Experimental results demonstrated that the pre-trained encoders (BanglaBert) achieved state-of-the-art performance. Among different prompting strategies, chain-of-thought (CoT) combined with few-shot prompting proved most effective. Following the HateSense framework, our system attained competitive micro-F1 scores: 0.741 (Task 1A), 0.724 (Task 1B), and 0.7233 (Task 1C). These findings affirm the effectiveness of transformer-based architectures for Bangla hate speech detection and suggest promising avenues for multi-task learning in low-resource languages.</abstract>
<identifier type="citekey">prama-etal-2025-computational</identifier>
<location>
<url>https://aclanthology.org/2025.banglalp-1.37/</url>
</location>
<part>
<date>2025-12</date>
<extent unit="page">
<start>430</start>
<end>442</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Computational Story Lab at BLP-2025 Task 1: HateSense: A Multi-Task Learning Framework for Comprehensive Hate Speech Identification using LLMs
%A Prama, Tabia Tanzin
%A Danforth, Christopher M.
%A Dodds, Peter
%Y Alam, Firoj
%Y Kar, Sudipta
%Y Chowdhury, Shammur Absar
%Y Hassan, Naeemul
%Y Prince, Enamul Hoque
%Y Tasnim, Mohiuddin
%Y Rony, Md Rashad Al Hasan
%Y Rahman, Md Tahmid Rahman
%S Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025)
%D 2025
%8 December
%I Association for Computational Linguistics
%C Mumbai, India
%@ 979-8-89176-314-2
%F prama-etal-2025-computational
%X This paper describes HateSense, our multi-task learning framework for the BLP 2025 shared task 1 on Bangla hate speech identification. The task requires not only detecting hate speech but also classifying its type, target, and severity. HateSense integrates binary and multi-label classifiers using both encoder- and decoder-based large language models (LLMs). We experimented with pre-trained encoder models (Bert based models), and decoder models like GPT-4.0, LLaMA 3.1 8B, and Gemma-2 9B. To address challenges such as class imbalance and the linguistic complexity of Bangla, we employed techniques like focal loss and odds ratio preference optimization (ORPO). Experimental results demonstrated that the pre-trained encoders (BanglaBert) achieved state-of-the-art performance. Among different prompting strategies, chain-of-thought (CoT) combined with few-shot prompting proved most effective. Following the HateSense framework, our system attained competitive micro-F1 scores: 0.741 (Task 1A), 0.724 (Task 1B), and 0.7233 (Task 1C). These findings affirm the effectiveness of transformer-based architectures for Bangla hate speech detection and suggest promising avenues for multi-task learning in low-resource languages.
%U https://aclanthology.org/2025.banglalp-1.37/
%P 430-442
Markdown (Informal)
[Computational Story Lab at BLP-2025 Task 1: HateSense: A Multi-Task Learning Framework for Comprehensive Hate Speech Identification using LLMs](https://aclanthology.org/2025.banglalp-1.37/) (Prama et al., BanglaLP 2025)
ACL