A Hybrid Architecture for Fake News Detection Using BERT Contextual and EmoBERTa Emotional Features

doi:10.22716/sckt.2025.13.4.015

All Issue

2025 Vol.13, Issue 4 Preview Page Next Page

Research Article

A Hybrid Architecture for Fake News Detection Using BERT Contextual and EmoBERTa Emotional Features 가짜뉴스 탐지를 위한 BERT 문맥 및 EmoBERTa 감정 특징 기반 하이브리드 아키텍처: Joon-Hwan Park¹, Min-Kyoung Shim², and Sujin Lee³
박준환¹, 심민경², 이수진³; ¹Undergraduate Student, Department of Artificial Intelligence and Data Science, Sejong University
²Undergraduate Student, Department of Artificial Intelligence, Sejong University
³Assistant professor, Department of Artificial Intelligence and Data Science, Sejong University

¹세종대학교 인공지능데이터사이언스학과 학부생
²세종대학교 인공지능학과 학부생
³세종대학교 인공지능데이터사이언스학과 조교수

31 December 2025. pp. 173-187

PDF

Abstract

가짜뉴스는 인간의 감정을 자극해 기존 뉴스보다 빠르게 확산하지만 이를 탐지하는 기존 연구는 주로 소수의 감정 범주를 사용하는 한계가 있었다. 본 연구는 가짜뉴스의 포괄적 감정을 반영하는 데 목적이 있다. GoEmotions의 28개의 정교한 감정 체계를 활용한 EmoBERTa를 이용하여 WELFake 데이터셋 텍스트의 ‘다차원 감정 특징’를 추출했다. 해당 감정 특징의 유효성에 대해 T-검정과 만-휘트니 U 검정을 진행하였고, 대부분이 가짜뉴스와 진짜 뉴스 집단 간 유의미한 통계적 차이를 보임을 입증했다. BERT-base-uncased로 추출한 문맥 벡터와 검증된 감정 벡터의 결합을 입력으로 하는 랜덤 포레스트 분류 모델을 제안했다. 제안 모델은 문맥 벡터만 사용한 모델에 비해 가짜 뉴스 탐지에서 성능 향상(Weighted F1-Score 0.14%p)을 보였다. 본 실험은 통계 연구와 함께 입증한 결과로 기초 연구를 넘어 사회과학 분야에 도구로 사용 가능성 의미를 갖는다.

Fake news tends to spread significantly faster than truthful news by stimulating human emotions; however, existing detection research has been limited by primarily utilizing a small number of basic emotional categories, failing to capture complex nuances. This study aims to overcome this limitation by reflecting the comprehensive and multidimensional emotional characteristics inherent in fake news. We specifically utilized EmoBERTa, fine-tuned on the GoEmotions dataset comprising 28 sophisticated emotional labels, to extract granular 'multidimensional emotional features' from the text of the WELFake benchmark dataset. The statistical validity of these extracted features was rigorously verified using T-tests and Mann-Whitney U tests, which proved that the majority of features showed significant differences between fake and real news groups. Finally, we proposed a Random Forest classification model that takes a combined input of contextual vectors extracted via the BERT-base-uncased model and the pre-validated multidimensional sentiment vectors. The proposed hybrid model demonstrated a performance improvement in fake news detection, achieving a Weighted F1-Score increase of 0.14 percentage points compared to the baseline model using only context vectors. This experiment holds significance beyond basic research; by providing statistical evidence of emotional distinctions, it suggests potential for practical application as a valuable tool in the field of social sciences.

Keywords

BERT

Affective computing

NLP

Fake news

Emotion recognition

Deep learning

Emotion classification

References

S. Vosoughi, D. Roy, and S. Aral, “The Spread of True and False News Online,” Science, Vol. 359, No. 6380, pp. 1146-1151, 2018.
10.1126/science.aap9559
Y. Chuai and J. Zhao, “Anger Can Make Fake News Viral Online,” Frontiers in Physics, Vol. 10, Article No. 970174, 2022.
10.3389/fphy.2022.970174
C. Martel, G. Pennycook, and D. G. Rand, “Reliance on Emotion Promotes Belief in Fake News,” Cognitive Research: Principles and Implications, Vol. 5, No. 1, p. 47, 2020.
10.1186/s41235-020-00252-3 >33026546 PMC7539247
S. Ippa, T. Okubo, and M. Hashimoto, “The Relationship Between Emotional and Other Factors in Information Diffusion,” IEEE Access, 2025.
10.1109/ACCESS.2025.3535547
J. Paschen, “Investigating the Emotional Appeal of Fake News Using Artificial Intelligence and Human Contributions,” Journal of Product & Brand Management, Vol. 29, No. 2, pp. 223-233, 2020.
10.1108/JPBM-12-2018-2179
P. Ekman, “An Argument for Basic Emotions,” Cognition & Emotion, Vol. 6, No. 3-4, pp. 169-200, 1992.
10.1080/02699939208411068
V. Kolev, G. Weiss, and G. Spanakis, “FOREAL: RoBERTa Model for Fake News Detection Based on Emotions,” Proceedings of ICAART (2), 2022.
10.5220/0010873900003116
D. Rozado, R. Hughes, and J. Halberstadt, “Longitudinal Analysis of Sentiment and Emotion in News Media Headlines Using Automated Labelling with Transformer Language Models,” PLOS One, Vol. 17, No. 10, e0276367, 2022.
10.1371/journal.pone.0276367 >36256658 PMC9578611
D. Demszky, et al., “GoEmotions: A Dataset of Fine-Grained Emotions,” arXiv preprint, arXiv:2005.00547, 2020.
10.18653/v1/2020.acl-main.372
J. Devlin, et al., “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” Proceedings of NAACL-HLT, pp. 4171-4186, 2019.
E. Kannan and L. A. Kothamasu, “Fine-Tuning BERT Based Approach for Multi-Class Sentiment Analysis on Twitter Emotion Data,” Ingénierie des Systèmes d’Information, Vol. 27, No. 1, 2022.
10.18280/isi.270111
C. Baek, J. Kang, and S. Choi, “Online Unstructured Data Analysis Models with KoBERT and Word2Vec: A Study on Sentiment Analysis of Public Opinion in Korean,” International Journal of Fuzzy Logic and Intelligent Systems, Vol. 23, No. 3, pp. 244-258, 2023.
10.5391/IJFIS.2023.23.3.244
A. R. Abas, et al., “BERT-CNN: A Deep Learning Model for Detecting Emotions from Text,” Computers, Materials & Continua, Vol. 71, No. 2, 2022.
10.32604/cmc.2022.021671
B. Farhoudinia, S. Ozturkcan, and N. Kasap, “Emotions Unveiled: Detecting COVID-19 Fake News on Social Media,” Humanities and Social Sciences Communications, Vol. 11, No. 1, 2024.
10.1057/s41599-024-03083-5
김백기, 장경배, “딥러닝과 감성사전을 결합한 하이브리드 감성분석 시스템 개발,” 사물인터넷융복합논문지, 제9권 제4호, pp. 21-31, 2023.
D. Hazarika, et al., “Conversational Transfer Learning for Emotion Recognition,” Information Fusion, Vol. 65, pp. 1-12, 2021.
10.1016/j.inffus.2020.06.005
T. Kim and P. Vossen, “EmoBERTa: Speaker-Aware Emotion Recognition in Conversation with RoBERTa,” arXiv preprint, arXiv:2108.12009, 2021.
P. K. Verma, et al., “WELFake: Word Embedding over Linguistic Features for Fake News Detection,” IEEE Transactions on Computational Social Systems, Vol. 8, No. 4, pp. 881-893, 2021.
10.1109/TCSS.2021.3068519

Information

Publisher :The Society of Convergence Knowledge
Publisher(Ko) :융복합지식학회
Journal Title :The Society of Convergence Knowledge Transactions
Journal Title(Ko) :융복합지식학회논문지
Volume : 13
No :4
Pages :173-187
DOI :https://doi.org/10.22716/sckt.2025.13.4.015

[1] S. Vosoughi, D. Roy, and S. Aral, “The Spread of True and False News Online,” Science, Vol. 359, No. 6380, pp. 1146-1151, 2018.
10.1126/science.aap9559

[2] Y. Chuai and J. Zhao, “Anger Can Make Fake News Viral Online,” Frontiers in Physics, Vol. 10, Article No. 970174, 2022.
10.3389/fphy.2022.970174

[3] C. Martel, G. Pennycook, and D. G. Rand, “Reliance on Emotion Promotes Belief in Fake News,” Cognitive Research: Principles and Implications, Vol. 5, No. 1, p. 47, 2020.
10.1186/s41235-020-00252-3 >33026546 PMC7539247

[4] S. Ippa, T. Okubo, and M. Hashimoto, “The Relationship Between Emotional and Other Factors in Information Diffusion,” IEEE Access, 2025.
10.1109/ACCESS.2025.3535547

[5] J. Paschen, “Investigating the Emotional Appeal of Fake News Using Artificial Intelligence and Human Contributions,” Journal of Product & Brand Management, Vol. 29, No. 2, pp. 223-233, 2020.
10.1108/JPBM-12-2018-2179

[6] P. Ekman, “An Argument for Basic Emotions,” Cognition & Emotion, Vol. 6, No. 3-4, pp. 169-200, 1992.
10.1080/02699939208411068

[7] V. Kolev, G. Weiss, and G. Spanakis, “FOREAL: RoBERTa Model for Fake News Detection Based on Emotions,” Proceedings of ICAART (2), 2022.
10.5220/0010873900003116

[8] D. Rozado, R. Hughes, and J. Halberstadt, “Longitudinal Analysis of Sentiment and Emotion in News Media Headlines Using Automated Labelling with Transformer Language Models,” PLOS One, Vol. 17, No. 10, e0276367, 2022.
10.1371/journal.pone.0276367 >36256658 PMC9578611

[9] D. Demszky, et al., “GoEmotions: A Dataset of Fine-Grained Emotions,” arXiv preprint, arXiv:2005.00547, 2020.
10.18653/v1/2020.acl-main.372

[10] J. Devlin, et al., “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” Proceedings of NAACL-HLT, pp. 4171-4186, 2019.

[11] E. Kannan and L. A. Kothamasu, “Fine-Tuning BERT Based Approach for Multi-Class Sentiment Analysis on Twitter Emotion Data,” Ingénierie des Systèmes d’Information, Vol. 27, No. 1, 2022.
10.18280/isi.270111

[12] C. Baek, J. Kang, and S. Choi, “Online Unstructured Data Analysis Models with KoBERT and Word2Vec: A Study on Sentiment Analysis of Public Opinion in Korean,” International Journal of Fuzzy Logic and Intelligent Systems, Vol. 23, No. 3, pp. 244-258, 2023.
10.5391/IJFIS.2023.23.3.244

[13] A. R. Abas, et al., “BERT-CNN: A Deep Learning Model for Detecting Emotions from Text,” Computers, Materials & Continua, Vol. 71, No. 2, 2022.
10.32604/cmc.2022.021671

[14] B. Farhoudinia, S. Ozturkcan, and N. Kasap, “Emotions Unveiled: Detecting COVID-19 Fake News on Social Media,” Humanities and Social Sciences Communications, Vol. 11, No. 1, 2024.
10.1057/s41599-024-03083-5

[15] 김백기, 장경배, “딥러닝과 감성사전을 결합한 하이브리드 감성분석 시스템 개발,” 사물인터넷융복합논문지, 제9권 제4호, pp. 21-31, 2023.

[16] D. Hazarika, et al., “Conversational Transfer Learning for Emotion Recognition,” Information Fusion, Vol. 65, pp. 1-12, 2021.
10.1016/j.inffus.2020.06.005

[17] T. Kim and P. Vossen, “EmoBERTa: Speaker-Aware Emotion Recognition in Conversation with RoBERTa,” arXiv preprint, arXiv:2108.12009, 2021.

[18] P. K. Verma, et al., “WELFake: Word Embedding over Linguistic Features for Fake News Detection,” IEEE Transactions on Computational Social Systems, Vol. 8, No. 4, pp. 881-893, 2021.
10.1109/TCSS.2021.3068519

The Society of Convergence Knowledge Transactions ISSN:2287-8920(Print) 융복합지식학회논문지

All Issue