Identifying Cyberbullying and Hate Speech via Sentiment Analysis of

Social Media Text Social Networks

Loveleen Kaur Pabla

, Prashant Kumar Jain

, Prabhat Patel

and Shailja Shukla

Department of Information Technology, Jabalpur engineering College, Jabalpur, Madhya Pradesh, India

Department of Electronics and Telecommunication, Rajiv Gandhi Proudyogiki Vishwavidyalaya, Bhopal, Madhya

Pradesh, India

Professor & Head, Department of Electrical Engineering, Jabalpur engineering College Jabalpur, Madhya Pradesh, India

Keywords: Sentiment Analysis, Text Mining, Natural Language Processing, Cyber Bulling, Hate Speech.

Abstract: In the current digital era, in addition to being a vital component of modern life, social media has also given

rise to the pervasive and dangerous issue of cyberbullying. The rise of social media platforms has coincided

with an increase in cyberbullying, a type of abuse that takes place on these sites. Social networks on the

internet are expanding quickly in many ways. These platforms are being utilised by various businesses to

advertise their goods and interact with final customers. Conversely, some groups are abusing these platforms

to spread hate speech, violent materials, and offensive or harassing material. The social climate both online

and offline is out of balance as a result of all these events. The proposed work analyses the text content of

social media posts, messages, and blogs to provide a research proposal on the identification of hate and

bullying content online. Machine learning methods and sentiment analysis are used in this context as crucial

instruments to handle large volumes of data. In addition to finding bullying terms, sentiment analysis has the

ability to identify victims in danger of hurting themselves or others. In order to detect signs of bullying in

social media posts, our research relies on applying deep learning and natural language comprehension

approaches. A Recurrent Neural Network with Long Short-Term Memory (LSTM) cells was constructed

using multiple embeddings. One method makes use of BERT embeddings, while the other uses OpenAI's

freshly published embeddings API in place of the embeddings layer.

1 INTRODUCTION

Hate speech and cyberbullying continue to occur in a

similar way to traditional forms. Direct bullying or

hate speech simply altered their appearance and had a

detrimental effect on the victim's mental state. This

social threat is becoming more widespread every day

on a variety of social media sites, not just in India but

globally as well. Teenagers are not the only age group

affected by these things; they work for all ages. It is

unacceptable to disseminate the dubious text by

various means of communication, such as social

media, SMS, emails, and others. Thus, text analysis

methods are helpful for examining texts and

identifying patterns of bullying and hatred. These

days, these methods are used for prediction,

recommendation, and decision making in a variety of

application domains, including business, banking,

education, engineering, and medicine. However,

assessing the attitudes in a document requires more

than just basic text analysis and categorization

methods. As a result, NLP-based text analysis is

employed to identify the attitudes and emotions

contained within the text. Understanding text

sentiments requires a solid understanding of natural

language processing (NLP). Human life preservation

may benefit from it. Finding text communication's

sentiment classes and intensity scores is the main goal

of the suggested effort. Finding the subclasses of

sentiment classes is the goal of the technique, which

offers more than just sentiment classification.

provides a sentiment score as well, which aids in

determining the degree of hate. This work focusses on

the domains of natural language processing (NLP)

and machine learning. NLP is a field that deals with

in-text blocks, text classification, and emotion

detection. Sentiment analysis can be used to content

found on forums, microblogs, e-commerce product

448

Pabla, L. K., Jain, P. K., Patel, P. and Shukla, S.

Identifying Cyberbullying and Hate Speech via Sentiment Analysis of Social Media Text Social Networks.

DOI: 10.5220/0013899800004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 3, pages

448-453

ISBN: 978-989-758-777-1

reviews, social media, and other sources. This effort

analyses text from a variety of public sources using

machine learning methods. Therefore, a suggestion to

build and implement a hate speech and cyberbullying

detection system is put forth. There have been

numerous contributions in this area recently from

different authors. However, the majority of them can

only forecast positive or negative orientations and

their subclasses. However, there are no appropriate

methods to address the influence of emotions.

Therefore, the suggested task entails creating a data

model to gauge how strongly the bully or hate speech

is conveyed in the text. We also attempted to

subdivide the emotion classes based on the feelings'

intensity. Social media has become an essential

component of everyday life in recent years, enabling

previously unheard-of levels of connection and

communication. But it has also played a part in the

increase in online harassment and cyberbullying

(Feinberg and Robey, 2009). Cyberbullying, in

contrast to traditional bullying, can occur at any time,

and because of the anonymity offered by the internet,

offenders are encouraged to behave without fear of

instant repercussions (Hasan et al., 2023). When

compared to in-person bullying, this anonymity

frequently results in higher rates of cyberbullying.

2 NATURAL LANGUAGE

PROCESSING

There have been numerous contributions in this area

recently from different authors. However, the

majority of them can only forecast positive or

negative orientations and their subclasses. However,

there are no appropriate methods to address the

influence of emotions. Therefore, the suggested task

entails creating a data model to gauge how strongly

the bully or hate speech is conveyed in the text. We

also attempted to subdivide the emotion classes based

on the feelings' intensity. Natural language describes

the speech and text that humans use to communicate

with one another. Text is all around us, and we use

menus, signage, emails, SMS, websites, and more.

We communicate via writing and speaking to one

another. Speaking might be simpler to learn than

writing. We connect with one another via text and

voice. Like other forms of data, natural language

requires tools for comprehension and reasoning.

 Automatic Summarization- Data is the

most valuable resource in this day and age.

We obtain the necessary and helpful

quantity of information from a variety of

sources. We have access to far more

knowledge than we can comprehend, and the

amount of information is overwhelming.

Since there will always be an abundance of

information available online, we require an

automatic text summarization system. The

process of producing a concise, precise

synopsis of text documents is known as text

summarization. In less time, text

summarization will provide us with

pertinent information. Automatic text

summarization relies heavily on natural

language processing.

 Question-answering- Question-answering

(QA) is another use case for natural

language processing. Although search

engines make the world's knowledge

accessible, they nevertheless fall short when

it comes to providing answers to our queries.

This is the route that businesses like Google

are taking. It is a branch of NLP and AI. Its

main goal is to create systems that can

automatically respond to human enquiries.

In order to create valid answers, a computer

system that comprehends the NL can

transform the sentences into an internal

representation. Syntax and semantic analysis

can produce the precise answers. Among the

difficulties NLP has in creating an effective

QA system are lexical gaps, ambiguity, and

multilingualism.

 Sentiment Analysis- Sentiment analysis is

another way that NLP is used. To determine

the sentiments of several posts, sentiment

analysis is utilised. When feelings are not

clearly conveyed, it is used to determine the

sentiment. Sentiment analysis is an NLP

application that businesses use to find out

what their customers think. Knowing what

their clients think of the goods and services

will be beneficial. They can use sentiment

analysis to assess their reputation based on

consumer reviews. Sentiment analysis helps

us better understand the motivations behind

the opinions expressed by analysing

sentiments in context.

Identifying Cyberbullying and Hate Speech via Sentiment Analysis of Social Media Text Social Networks

449

3 HATE SPEECH AND CYBER

BULLING

Targeting an individual or a group on the basis of their

religion, race, ethnic origin, handicap, sexual

orientation, or gender is known as online hate speech,

and it typically occurs on social media. It is where

several tensions converge: [weasel words]. It is vivid

when diverse groups express themselves differently

from one another. Multilateral and multi-stakeholder

procedures is a broad word. Hate speech is becoming

a common term and is growing daily. Threats to the

security of both individuals and groups are being

mixed together. Google, Twitter, Facebook, and other

social media sites started defining hate speech in their

own ways. It is necessary for national and regional

bodies to raise awareness of such phrase. Because of

the speed and breadth of the Internet, governments

find it difficult to enforce laws in this situation.

Because of the problems with hate speech, private

channels for expression that operate in public spaces,

like Facebook and Twitter, have become more

popular. What is hate speech, how it relates to offline

and success, what constitutes apparent hate speech,

and likely punishment have led to a lot of discussion

about the solution and how it should be based.

However, this emphasis has also hindered more

thorough efforts to identify the root reasons of the

phenomena.

Bullying or harassment via digital devices, such

as smartphones, laptops, and computers, is known as

cyberbullying. Chat rooms, social networking, and

gaming are examples of platforms where

cyberbullying can happen since users can read and

share content. The various forms of bullying include

abusive remarks, text messages, and messages. It

includes sending, posting, and disseminating

inaccurate or unfavourable information. Do you think

your child has a smartphone addiction? Has he or she

grown more reclusive and socially insensitive?

Changes in a child's behaviour might cause parents

great concern. Watch out! It's possible that your

youngster is being bullied online. Parents can report

cyberbullying according to research on the topic.

Some typical forms of cyberbullying are as follows:

 Posting offensive or derogatory remarks

 rumours, or pictures about someone

 making a phoney or offensive webpage

about someone

 threatening or inciting someone online

 inciting hate speech about someone based on

their race, religion, ethnicity, or politics

 posing as someone else online in order to

post personal or false information about

someone

Information about someone's social media accounts is

obtained and utilised for online abuse, slander, and

other purposes. Cyberbullying has seen a sharp

increase in India as social media and reasonably

priced broadband services have become more widely

available. Studies show that over eight out of ten

people experience some form of cyberbullying. Of

these, around 63% were subjected to online

harassment and abuse, and 59% were the target of

unfounded rumours and gossip. Furthermore, a

number of research show that India has the greatest

rate of cyberbullying in Asia. In particular, 50% of

Indian city women experience abuse online.

4 PROPOSED WORK AND

PROPOSED DATA MODEL

The proposed work's goal is to research and enhance

the sentiment analysis method for identifying hate

speech and posts related to cyberbullying on social

media. Anyone can post anything on social media.

Therefore, the following goals are set in order to keep

an eye on and maintain the cleanliness of social

networks.

 To investigate and learn about hate

speech and cyberbullying detection

methods- A survey of the current methods

and resources for sentiment analysis, hate

speech, and cyberbullying is necessary

because there has been a lot of work done

recently on social media text processing for

sentiment label recovery. As a result, this

survey aids in the formulation of our

solution development plan.

 For automatic sentiment identification, to

expand and enhance current sentiment

analysis methods- The promising

classification methods are retrieved from the

literature review in order to improve their

performance in sentiment classification.

Furthermore, the updated features allow for

precise detection of hate speech and

bullying.

 Incorporating sentiment's intensity and

semantics into text- Here, the sentiment

analysis method that was previously

provided is changed to determine the

sentiment impact on the target individual or

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

450

community. As a result, a scale and score are

created to quantify the hate and bad attitudes

expressed in the text.

 To research and evaluate the suggested

cyberhate detection framework's

performance in comparison to cutting-

edge methods- The rationale behind the

suggested model is attempted to be

illustrated. Furthermore, a performance

comparison between the suggested work and

pertinent existing methodologies is shown.

Additionally, utilising the information and

facts that were available on Twitter, we

attempted to follow an actual case.

Figure 1: Proposed Data Model.

Figure 1 shows the proposed data model. The goal

of the proposed work is to enhance the sentiment

analysis method already in use. A score system based

on delivered text will be incorporated into the

extension to gauge the sentiment intensity.

Cyberbullying and hate speech content detection

applications are utilised in conjunction with this

scoring system. Thus, image 1 illustrates a basic

model. Twitter is therefore chosen as the social media

platform for data collection and experimentation. In

this context, we can locate a variety of offline and

online dataset resources. Furthermore, social media

sites like Facebook and Twitter offer web APIs for

retrieving data from live feeds. We must enhance the

quality of the raw offloaded data after offloading the

necessary data. Thus, in order to improve the quality

of the data, pre-processing is used first. Since the

offloaded data is essentially in text format, or

unstructured format, it comprises a variety of

undesirable characters and stop words. The offline

contents were attempted to be removed by the pre-

process. Following content refining, feature

extraction methods are applied. Here, we must

identify a few text-based or natural language

processing (NLP) feature selection techniques and

assess their efficacy before implementing them in the

feature extraction of the suggested model. To identify

the desired patterns, supervised, semi-supervised, and

unsupervised learning approaches are used with the

computed features from the pre-processed text data.

This procedure is referred to as the suggested system's

training. Lastly, the sentiment class for the provided

data is generated by the suggested model. In essence,

we are unable to judge a social media post's criticality

based on its polarity, or whether it is accurate or not.

In order to determine the severity and negativity of a

hate speech post, we are attempting to calculate a

sensitivity score utilizing the various facts that are

currently accessible. An overview of the suggested

data model is presented in this paper, and the specifics

of the model are provided in the next piece.

5 EXPERIMENTATION

Binary Cross Entropy Loss (BCELoss), which works

well for binary classification tasks, is used to train

the model. The definition of the BCELoss is

𝐿



𝑦log



𝑝







1𝑦



log



1𝑝



(1)

Where, the anticipated probability for the positive

class is denoted by p and the true label by y. The

classifier performs better when BERT-base or

OpenAI embeddings that is our hybrid architectures

are used than when the RNN is used alone.

Identifying Cyberbullying and Hate Speech via Sentiment Analysis of Social Media Text Social Networks

451

Figure 2: Bert/Openai Api Embeddings in an Rnn Network.

Providing high-quality feature vectors that were

taken from the text data allows for this improvement.

The RNN network with BERT or OpenAI

embeddings is displayed in Figure 2.

Implementation Details- Training and testing

data were divided in an 80:20 ratio. We employed

dropout to reduce overfitting and gradient clipping to

avoid exploding gradients, with a learning rate of

0.001. The training procedure, which lasted ten

epochs, used the Adam optimiser.

6 CONCLUSIONS

Bullying and hate are becoming more and more of a

social threat worldwide, manifesting in different ways

such as race, religion, body, physical ability, and

others. It is overly delicate and has a detrimental

effect on a person's thinking as well as the target

community. Furthermore, hatemongers are using

digital communication platforms and social media to

spread their malicious content. In this regard, the

suggested work is an endeavour that aids in the

collection and examination of textual content from

widely used social media platforms in order to

identify contents that are counterfeit. The suggested

effort first identifies the questionable content,

preventing it from going viral across many channels

and ultimately saving many lives.

 To accomplish the necessary goal, the

following next stages are included in the

suggested sentiment analysis based on

bullying and hate speech detection

technique:

 To identify appropriate sentiment-based

feature selection methods and text feature

selection methods, do a comparative

performance analysis.

 To handle the text input and pertinent feature

set, find a classifier that is both accurate and

efficient.

 Look for some engineered qualities that

contribute to the legitimacy of the

source.For a target event, identify the origin

of compromised content and potential

distribution channels.

Our research showed that in capturing the nuances

of cyberbullying language, both OpenAI and BERT

embeddings performed noticeably better than a

simple RNN model. OpenAI embeddings were the

most effective of them all, demonstrating their

exceptional capacity to comprehend sentiment and

context in user-generated information. Going

forward, the dataset's modest size presents difficulties

for model performance. In order to solve this, we

intend to investigate zero-shot or few-shot learning

techniques in subsequent research, which may

improve the model's generalisation and precision in

forecasting cases of cyberbullying with a small

amount of labelled data.

REFERENCES

A. kumar, “What is data mining”, CSE Dept, Dr. APJ Abdul

Kalam UIT Jhabua (M.P.) “Investment: Unit – 1,

Security Analysis and Portfolio Management”,

http://www.pondiuni.edu.in/storage/dde/downloads/

finiv_ sapm.pdf

B. Krawczyk, “Learning from imbalanced data: open

challenges and future directions”, Prog Artif Intell

(2016) 5:221–232

C. Clavel, Z. Callejas, “Sentiment Analysis: From Opinion

Mining to Human-Agent Interaction”, IEEE

Transactions on Affective Computing, VOL. 6, NO. X,

XXXXX 2015 “Preparing for the Future of Artificial

Intelligence”, Executive Office of the President

National Science and Technology Council Committee

on Technology, Oct. 2016, https:// obamawhitehouse.

archives.gov/sites/default/files/whitehouse_files/micro

sites/ostp/NSTC/preparing_ for_the_future_of_ai.pdf

D Cross. 2008. Cyberbullying: International comparisons,

implications, and recommendations. In 20th biennial

meeting of the International Society for the Study of

Behavioural Development, Wurzburg, Germany.

G. W. Blood, I. M. Blood, “Cyberbullying: Responsibility,

Concerns and Personal Experiences of School-based

Speech-Language Pathologists”, International Journal

for Infonomics (IJI), Volume 9, Issue 1, March 2016

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

452

G. Peters, R. Weber, “DCC: a framework for dynamic

granular clustering”, Granul. Comput. (2016) 1:1–11,

DOI 10.1007/s41066-015-0012-z

I. V. Serban, C. Sankar, M. Germain, S. Zhang, Z. Lin, S.

Subramanian, T. Kim, M. Pieper, S. Chandar, N. R.

Ke,S. Rajeshwar, A. d. Brebisson, J. M. R. Sotelo, D.

Suhubdy, V. Michalski, A. Nguyen, J. Pineau, Y.

Bengio, “A Deep Reinforcement Learning Chatbot”,

arXiv:1709.02349v2 [cs.CL] 5 Nov 2017

J. Wan, D. Wang, S. C.H. Hoi, P. Wu, J. Zhu, Y. Zhang, J.

Li, “Deep Learning for Content-Based Image Retrieval:

A Comprehensive Study”, Proceedings of the ACM

International Conference on Multimedia: November 3-

7, 2014, Orlando. pp. 157-166

J. Jayasree, Sri M. A. Mathi, S. Malini, E. Bavithra, Dr. P.

Boobalan, “Product Ranking System in E-Commerce

Website for Validation using Sentimental Analysis”,

International Journal of Science Technology &

Engineering, Volume 3, Issue 09, March 2017

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina

Toutanova. 2019. Bert: Pre-training of deep

bidirectional transformers for language understanding.

In North American Chapter of the Association for

Computational Linguistics.

Jalal Omer Atoum. 2020. Cyberbullying detection through

sentiment analysis. 2020 International Conference on

Computational Science and Computational Intelligence

(CSCI), pages 292–297.

Jalal Omer Atoum. 2021. Cyberbullying detection neural

networks using sentiment analysis. 2021 International

Conference on Computational Science and

Computational Intelligence (CSCI), pages 158–164.

M. Rajan, T. S. Rinku, V. Bhojane, “Information Retrieval

in Malayalam Using Natural Language Processing”,

International Journal of Scientific & Engineering

Research, Volume 5, Issue 6, June-2014

M. M. Kampert, J. J. Meulman, J. H. Friedman, “rCOSA:

A Software Package for Clustering Objects on Subsets

of Attributes”, Journal of Classification 34:514-547

(2017)

M. M. Mironczuk, J. Protasiewicz, “A recent overview of

the state-of-the-art elements of text classification”

,Expert Systems With Applications, 106, 2018, 36-54

Michael Agbaje and Oreoluwa Afolabi. 2024. Neural

network-based cyber-bullying and cyber-aggression

detection using twitter(x) text. Revue d’Intelligence

Artificielle.

Nhan Cach Dang, María N. Moreno García, and Fernando

de la Prieta. 2020. Sentiment analysis based on deep

learning: A comparative study. ArXiv,abs/2006.03541.

S. Zhang, C. Zhang, Q. Yang, “Data Preparation For Data

Mining”, Applied Artificial Intelligence, 17:375– 381,

2003, Copyright # 2003 Taylor & Francis, 0883-

9514/03$12.00+.00,DOI:10.1080/0883951039021926

“DataMiningOverview”,https://www.tutorialspoint.co

m/data_mining/dm_overvie w.htm

S. Khemka, “Why Data Mining? Editor’s Point of View”,

digital valley, Vol.1, No 4, July 2018

Stefano Baccianella, Andrea Esuli, Fabrizio Sebastiani, et

al. 2010. Sentiwordnet 3.0: an enhanced lexical

resource for sentiment analysis and opinion mining. In

Lrec, volume 10, pages 2200–2204. Valletta.

T. Wuest, D. Weimer, C. Irgens, K. D. Thoben, “Machine

learning in manufacturing: advantages, challenges, and

applications”, Production & Manufacturing Research:

An Open Access Journal, 2016, VOL. 4, NO. 1, 23–45

U. Fayyad, G. P. Shapiro, P. Smyth, “From Data Mining to

Knowledge Discovery in Databases”, AI Magazine

V. Gupta, G. S. Lehal, “A Survey of Text Mining

Techniques and Applications”, Journal of Emerging

Technologies in Web Intelligence, Vol. 1, No. 1, Aug

2009

W. Medhat, A. Hassan, H. Korashy, “Sentiment analysis

algorithms and applications: A survey”, Ain Shams

Engineering Journal 2014, 5, 1093-1113

Y. Roh, G. Heo, S. E. Whang, “A Survey on Data

Collection for Machine Learning”,arXiv:1811.03402v2

[cs.LG] 12 Aug 2019

Identifying Cyberbullying and Hate Speech via Sentiment Analysis of Social Media Text Social Networks

453