STRUKTURA RADA

THE PROBLEM OF HARMFUL CONTENT ON SOCIAL NETWORKS

WHY IDENTIFYING HARMFUL CONTENT IS A SERIOUS TOPIC TODAY

- how the spread of harmful content, such as extreme political opinions, affects user behavior.

SOCIAL NETWORKS AS A SECOND HOME FOR YOUNG PEOPLE

CONSEQUENCES OF HARMFUL CONTENT harms caused by hate speech

TECHNICAL METHODS FOR IDENTIFYING HARMFUL CONTENT

APPROACHES TO PROBLEM SOLVING

CONCLUSION AND DIRECTIONS FOR FUTURE RESEARCH

1. Meggyesfalvi, B. (2021.) Controlling Harmful Online Content on Social Media Platforms, Internal
Affairs Review 69 (6), 26-38. <[Link] visited 15.11.2024.

This article seeks to identify the challenges faced when policing online harmful material in social
media platforms such as who is responsible for managing harmful online content (legal and illegal) or
how to reduce the appearance of harmful content on social media platforms, and how to create
better conditions for the people handling the created harmful content to work effectively. The
context of policing online harmful material in social media platforms was analysed from different
perspectives. This paper is useful for my research as it deals with how the control of harmful online
content is implemented, and how these processes could be improved which is one of the key points
of my research.

2. Seble, H., Muluken, S., Edemealem, D., Kafte, T., Terefe, F., Mekashaw, G., Abiyot, B. and Senait,
T (2023). Hate speech detection using machine learning: a survey, Academy Journal of Science
and Engineering 17(1) 88-109.
<[Link]
ACHINE_LEARNING_A_SURVEY> visited 15.11.2024.

In this paper authors have made a systematic review of the advantages and disadvantages of a variety
of methods on the automatic detection of hate speech. Studies conducted in various techniques were
reviewed and [Link] techniques are keywordbased, machine learning, deep learning, and
hybrid techniques. Authors explore the challenges associated with labeling, dataset imbalance, and
the nuances of context in hate speech detection. Additionally, future research directions are
discussed in detail. This paper is useful for my research as it provides a systematic review of literature
in this field, with a focus on techniques and other state-of-the-art technologies with their gaps and
challenges which is start point for my research on the given topic.
 3. Subramanian, M., Easwaramoorthy Sathiskumar, V., Deepalakshmi, G., Cho, J., & Manikandan,
G. (2023). A survey on hate speech detection and sentiment analysis using machine learning
and deep learning models. Alexandria Engineering Journal, 80, 110-121.
<[Link] visited 16.11.2024.

This survey article provides a comprehensive overview of recent advancements in hate speech
detection and sentiment analysis on online content using machine learning and deep learning
models. The authors examine the strengths and weaknesses of different models and methods, and
how well they detect hate speech. Finally, they outline areas where more study is needed and
suggest potential new avenues for exploration in the field of hate speech identification and sentiment
analysis. The relevance of this paper for my research is that it covers two important topics, the
categorization of hate speech according to its linguistic characteristics and its application and the
accurate identification and classification of hate speech and sentiment in online text. Additionally, it
emphasizes the role of datasets, and challenges in data labeling, and points out gaps in research with
potential directions for future researchers.

4. Schmidt, A., Wiegand, M. (2017.). A Survey on Hate Speech Detection using Natural Language
Processing. Proceedings of the 5th International Conference on Natural Language Processing
(NLP), 1-10. < [Link] visited 17.11.2024.

This article provides a comprehensive overview of the methods and techniques used in the detection
of hate speech in text using Natural Language Processing (NLP). It provides a short, comprehensive
and structured overview of automatic hate speech detection, and outlines the existing approaches in
a systematic manner, focusing on feature extraction in particular.

It is relevant because it is mainly aimed at NLP researchers who are new to the field of hate speech
detection and aim to explore more about machine learning applications for online safety, content
filtering, and social media moderation with aim to understand the techical aspects of hate speech
detection and the ethical dilemmas that arise in the process.

5. Parihar, A., Thapa, S., & Mishra, S. (2021). Hate Speech Detection Using Natural Language
Processing: Applications and Challenges. Proceedings of the 2021 5th International Conference
on Optical and Electronic Imaging (ICOEI), 1302-1308.
<[Link] visited 17.11.2024.

In this paper, authors discuss the relevant works done in the field of hate speech detection offering a
clear overview of the current state of hate speech detection research. This paper provides an in-
depth examination of the use of Natural Language Processing (NLP) for hate speech detection,
focusing on both the applications and challenges in the field. Different types of hate speech like
racism, sexism, religious hate speech, etc., and the various methods proposed to tackle them are
discussed. The authors conclude the paper with necessary suggestions and the works that need to be
carried out in the future. This paper offers a discussion about the ways in which machine learning and
deep learning are used to control hate speech describes the related works in this domain along with
the challenges and presents possible solutions to tackle the challenges. Hence, the content of this
paper is relevant to my research.

6. Davidson, T., Warmsley D., Macy, M., Weber, I. (2017.) Automated Hate Speech Detection and
the Problem of Offensive Language, Proceedings of the Eleventh International AAAI Conference
on Web and Social Media , 11 < [Link] > visited
13.11.2024.

In this article, the authors are investigating the distinction between hate speech and offensive
language for a better understanding of the challenges associated with the automated detection of
hate speech on social media platforms. Using a large dataset of Twitter posts, authors develop
machine learning classifiers to examine text and categorize tweets into three groups: hate speech,
offensive language, or neither. Furthermore, they analyze the results in order to better understand
how we can differentiate between them and provide advice on how to strengthen moderation
protocols and guide legislation to combat hate speech online. This essay addresses the challenge of
identifying and classifying hate speech and offensive language, as well as the challenges of
differentiating between complex contexts, which makes it useful to my research. The research is well-
supported by experiments and offers practical suggestions for improving machine learning models in
this area.

7. Gongane, V.U., Munot, M.V. & Anuse, A.D. (2022). Detection and moderation of detrimental
content on social media platforms: current status and future directions. Social Network Analysis
and Mining, 12 (129) [Link]

This research paper provides a comprehensive review of the current methods used for detecting and
moderating harmful content on social media platforms. The authors examine reported detrimental
contents such as fake news, rumors, hate speech, aggressive, and cyberbullying which raise up as a
major concern in the society and detecting and moderating such content is a prime need of time. So,
AI-based methods like Natural Language Processing (NLP) with Machine Learning (ML) algorithms and
Deep Neural Networks is rigorously deployed for detection and moderation of detrimental content on
social media platforms. Authors analyze and assess the existing approaches to automatically
detecting and moderating such content and also addresses the ethical implications of content
moderation, including issues related to freedom of expression, bias, and the potential for over-
censorship. The authors conclude by suggesting a number of future directions for enhancing
automated content moderation systems. These include utilizing more sophisticated machine learning
models, incorporating multimodal data (such as text, images, and videos), and making moderation
algorithms more transparent and explainable.
 8. Kiritchenko, S., Nejadgholi, I., Fraser K. C. (2021.) Confronting Abusive Language Online: A
Survey from the Ethical and Human Rights Perspective, Journal of Artificial Intelligence Research
71 431-478, <[Link] visited 15.11.2024.

The article provides a review of a large body of NLP research on automatic abuse detection with a
new focus on ethical challenges, organized around eight established ethical principles: privacy,
accountability, safety and security, transparency and explainability, fairness and non-discrimination,
human control of technology, professional responsibility, and promotion of human values. The
authors examine the task of automated abusive language detection from the ethical viewpoint,
bringing together technical and social issues under a single ethical and human rights framework.
Additionally, they highlight the need to examine the broad social impacts of technology and to bring
ethical and human rights considerations to every stage of the application life-cycle. This article is a
beneficial source for my research since it provides information about ways to regulate online abuse
while respecting users' civil liberties, information about the intersection of digital policy, human
rights, and social media ethics, and insights into understanding the ethical implications of using
automated systems and AI for detecting and moderating harmful speech.

9. Chung, Y.-L., Kuzmenko, E., Tekiroglu, S. S., & Guerini, M. (2019). CONAN - COunter NArratives
through Nichesourcing: A multilingual dataset of responses to fight online hate speech.
Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2819–
2829. Association for Computational Linguistics. <[Link]
visited 17.11.2024.

In this paper was described CONAN: the first large-scale, multilingual, and expert-based hate
speech/counter-narrative dataset for English, French, and Italian. Together with the collected data it
also provides several types of metadata: expert demographics, hate speech sub-topic, and counter-
narrative type. Finally, the authors expanded the dataset through translation and paraphrasing. The
authors focus on the idea of "niche sourcing" – gathering responses to hate speech that not only
counteract harmful messages but also provide alternative, positive perspectives in various languages
and cultural contexts. Additionally it provides an aproach which is detecting and managing online
hate speech by shifting the focus from purely identifying and removing harmful content to fostering
positive, alternative narratives. Additionally, the authors emphasize the potential of this dataset to
improve automated systems for detecting hate speech and generating responses that are
contextually appropriate and linguistically accurate. As it provides alternative solutions and adequate
responses in terms of laws and policies regarding hate content on social media platforms, I find this
article helpful for my research.

10. Chuyi Sheng, Automated Content Moderation, 6 Geo. L. Tech. Rev. 351 (2022).
<[Link]
[Link] > visited 13.11.2024.

In this review author has reviewed in depth technological, legal, and ethical issues that arise
when algorithms are put to work to monitor and regulate user-generated content. The content
moderation mechanism is understood by analyzing: the primary organizational structures for the
moderation, when the moderation happens, how transparent and open are the underlying rules of
the moderation; and who is in charge. The review critiques the effectiveness of automated tools in
addressing harmful or illegal content, pointing at concerns regarding transparency, accountability, and
bias. Author discusses various regulatory schemes with recomendations for
a better balance between liberty of expression and content moderation. The review provides
summary of keypoint definitions and analyse use of automated systems in content
moderation in digital platforms and presents challenges and implications of the use of automated
systems in content moderation in digital platforms.

11. Yousaf, K., & Nawaz, T. (2022). A deep learning-based approach for inappropriate content
detection and classification of YouTube videos. IEEE Access, 10, 16283–16298.
[Link]

In this study, a novel deep learning-based architecture is proposed for the detection and classification
of inappropriate content in videos with special focus towards young viewers and their safety. For this,
the proposed framework employs an ImageNet pre-trained convolutional neural network (CNN)
model to extract video descriptors, which are then fed to a bidirectional long short-term memory
(BiLSTM) network to learn effective video representations and perform multiclass video classification.
An attention mechanism is also integrated after BiLSTM to apply attention probability distribution in
the network. The study emphasizes textual, visual, and auditory features to identify harmful or
unsuitable material. Authors targeted two types of objectionable content geared towards young
viewers, one, which contains violence, and the second, which includes sexual nudity connotations.
video platform management and designing automated systems for content curation. It can assist any
video sharing platform to either remove unsafe video or blur/hide any portion of video involving
unsafe content. Secondly, it may also help in the development of parental control solutions on the
web via plugins or browser extensions where children inappropriate content filters automatically. As
it stands out for its multi-modal approach, integrating video frames, audio transcripts, and metadata
for enhanced accuracy this article is particularly relevant for my research.

12. Sheth, A. P., Shalin, V. L., & Kursuncu, U. (2021). Defining and detecting toxicity on social
media: Context and knowledge are key. arXiv. <[Link]
visited 15.11.2024.

In this paper, authors define and critically examine toxicity, various types of toxic behavior, including
hate speech, bullying, and misinformation, to provide a foundation drawing social theories, provide
an approach that identifies multiple dimensions of toxicity and incorporates explicit knowledge in a
statistical learning algorithm to resolve ambiguity across such dimensions. Specifically, authors
highlighted the significance of multi-level analysis of data, namely, content, individual and
community, and the features necessary to determine toxicity.
They put a focus on enhancing the accuracy and fairness in content moderation
systems using context-aware methods based on structured knowledge graphs.

This paper identifies three issues: identify the psychological and social dimensions of the problem,
the limitations of contemporary computational approaches, and outline an advanced technical
approach founded on knowledge-driven context based analysis. The authors are highlighting gaps in
current AI models by emphasize context and cultural sensitivity which is useful direction for building
responsible and inclusive AI technologies.

13. Santos, F. C. C. (2023). Artificial intelligence in automated detection of disinformation:

This is an article discussing how AI has been used to analyze and fight against disinformation and
provides a thematic evaluation of the methodologies, technologies, and challenges that accompany
these works. While this article examines different approaches to leverage AI in combating
disinformation, the main focus is on automated fact-checking, which has been supported by existing
research. The author covers a range of approaches and gives review of methods for detection of
disinformation through language and sentiment analysis, using human-in-the-loop AI systems and the
application of AI blockchain in the automated detection of disinformation. Conclusions present the
advances in AI that are geared towards more effective disinformation detection and also the
challenges such as algorithmic and data biases. This resource is useful in contextualizing the balance
between technology and media ethics behind such systems in place and because it explores how the
combination of blockchain and AI technologies can be used to automate the process of
disinformation detection.

14. De Cock Buning, M. (2018.), A multi-dimensional approach to disinformation : report of the

15. Singh, Vivek K., Isha Ghosh, and Darshan Sonagara. 2021. Detecting fake news stories via
multimodal analysis. Journal of the Association for Information Science and Technology 72: 3–
17

16. Demartini, Gianluca, Stefano Mizzaro, and Damiano Spina. 2020. Human-in-the-loop Artificial
Intelligence for Fighting Online Misinformation: Challenges and Opportunities. IEEE Data
Engineering Bulletin 43: 65–74.

17. Kertysova, Katarina. 2018. Artificial intelligence and disinformation: How AI changes the way
disinformation is produced, disseminated, and can be countered. Security and Human Rights
29: 55–81.

18. Nakov, Preslav, David Corney, Maram Hasanain, Firoj Alam, Tamer Elsayed, Alberto Barrón-
Cedeño, Paolo Papotti, Shaden Shaar, and Giovanni Da San Martino. 2021. Automated fact-
checking for assisting human fact-checkers. arXiv arXiv:2103.07769.
19. Truică, C.-O., Constantinescu, A.-T., & Apostol, E.-S. (2017.). STOPHC: A harmful content
detection and mitigation architecture for social media platforms. Proceedings of the Eleventh
International AAAI Conference on Web and Social Media (ICWSM 2017) , 512-515,
[Link] visited 13.11.2024.

Multimodal approaches

Challenges

Impact on social media hate speech