Investigating Drug Trafficking Using Encrypted Messengers: NLP

and Data Analysis Approaches in Cybersecurity

Rushil Gautam, Kartik, Shravan Singh, Sarthak Srivastava and Dhanshri Parihar

Department of Computer Science and Engineering, Ajay Kumar Garg Engineering College, Ghaziabad, Uttar Pradesh,

India

Keywords: Cybersecurity, Drug Trafficking, Encrypted Messaging, NLP, Dark Web, Digital Forensics.

Abstract: The rise of encrypted messaging platforms has given the dealers of illegal drugs a new channel for their

transactions, making it difficult for law enforcement authorities to deal with them. The paper studies usage of

Natural Language Processing (NLP)and data analysis tech to find and analyze drug trafficking activities from

imposts on encrypted messaging platforms like WhatsApp and Telegram. By an analysis of digital forensic

operations together with sophisticated machine learning models, this research is directed at finding crime

patterns, retrieving temporal digital traces that have been erased and even building proposals for

countermeasures in order to mitigate the cybersecurity risks connected with online drug selling.

1 INTRODUCTION

The rise of the encrypted messaging services has

brought a revolution to the way of communication,

and it has provided users with privacy and security.

Nevertheless, this network has also become a center

for criminal acts among them drug trafficking.

However, the TB-Drug Test-Plus® was not

successful in finding a creating value-added

diagnostics-based TB pharm Dx which was later

licensed by DiaSorin. It’s a versatile test that gives

results for the common types of TB and at the same

time there’s also to people at risk of TB who do not

use this test correctly. Untraditionally, more cases of

TB have been ending up at hospitals where aches are

commonprecursorofdiagnosticsfollowedbybriefhospi

tal’sIdentifyapplicablefundingagencyhere.Ifnone,

delete this. stay instead of going to IHC because IHC

specialists were not skilled in linking the two. Unlike

traditional drug markets, online transactions on

platforms like WhatsApp, Telegram, and Signal

leverage encryption to evade detection, making it

increasingly difficult for law enforcement to track and

intercept these activities. This work is being aim edat

contributing with theoretical frameworks

(providing a comprehend), tending and cowitch(dete

ction) and other(identification) of them drug trafficki

ng issues on encrypted messaging platforms.

2 USECASES

2.1 Law Enforcement

Citizens have been alerted to hidden drug deals by

chatting and talking in code in the digital sphere! In

real time, the system can identify and interrupt

trafficking networks faster, if used properly.

2.2 Academic Research

With this platform, scientists can research the trends

of traffickers, follow the development of the slang,

and learn about the traffic network without end

angering the private lives of individuals.

2.3 Public Policy Formulation

Utilizing insights from this application method assists

the government in perceiving and drawing data-

driven actions and policies to fight drug trafficking.

the figure 1 shows the: Most Used Drugs.

Gautam, R., Kartik, , Singh, S., Srivastava, S. and Parihar, D.

Investigating Drug Trafﬁcking Using Encrypted Messengers: NLP and Data Analysis Approaches in Cybersecurity.

DOI: 10.5220/0013943100004919

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 5, pages

743-750

ISBN: 978-989-758-777-1

743

Figure 1: Most Used Drugs.

3 DIFFERENT NLP MODELS

FOR DRUG TRAFFIC KING

DETECTION

Some NLP approaches have been employed to

effectively identify drug trafficking activity. This

section presents some important models and their

descriptions, applications, and potential future

directions.

3.1 BERT – Based Models

Description: BERT (Bidirectional Encoder

Representations from Transformers) is a pre-trained

transformer-based model widely known for its

context-aware understanding of the con- textual

meaning of words in a sentence.

Application: BERT is applied in the analysis of

encrypted chat logs, social media posts and online

conversations to detect drug trafficking and

especially the detection of slang, euphemisms,

contextually relevant and so-called contextual words

that are commonly used by traffickers.

Advantages: BERT is highly accurate when

classified as text, and able to learn from and adapt to

changing language patterns. That makes BERT an

ideal tool in the fight against drug-related activity

online.

3.2 Graph Neural Networks (GNNs)

Description: GNNer is used to represent

relationships between different” devices” such as

hashtags, users and posts such as nodes and

Edges in a graph structure.

Application: Such networks can help identify human

trafficking by extracting correlations between users,

given keywords and content shared on social

networking sites such as Twitter and Instagram.

Advantages: GNNS has excellent ability to find

hidden correlations and intricate conditions in a large

-scale data set, making them very suitable for network

analysis.

3.3 Large Language Models (LLMs)

Description: Large Language Models (LLMs)such

as Chat- GPT and GPT-4 use knowledge-informed

prompts to analyze text data efficiently. Application:

These models are employed for detecting drug

trafficking activities by understanding and analyzing

deceptive language and evolving terminologies.

Advantages: LLMs excel in handling class-

imbalanced datasets and discovering new patterns

with limited labeled data, making them highly

adaptive.

3.4 Heterogeneous Graph Prompt

Learning (LLM-Het GDT)

Description: LLM-Het GDT combines Large

Language Models (LLMs) with Heterogeneous

Graph Neural Networks (HGNNs)to improve

detection accuracy, particularly in class imbalanced

scenarios.

Application: The system analyzes interactions

between users, posts, and keywords to detect drug

trafficking activities on platforms like Twitter.

Advantages: This approach is efficient, scalable, and

capable of addressing issues related to label scarcity

and data imbalance.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

744

3.5 Deep Learning Models for Image

and Text Analysis

Description: They use a combination of image

processing and NLP approaches to analyze

multimodal data from social media.

Application: They detect drugs in images while also

matching captions and hashtags to see where they ’re

being used. The figure 2 shows the NLP Models

Comparison.

Advantages: By applying is approach to multiple

data sources it can be said that they provide a broad

and comprehensive solution for identifying patterns

in drug trafficking in relation to various content.

Figure 2: NLP Models Comparison.

3.5.1 Summary

The above we can see different way sin which models

of NLP and Deep Learning are used to effectively

fight drug trafficking; Based on the more complex

strategy, these models not only recognize suspicious

activities, but also consider the constantly changing

language and the behavior of smugglers. This section

presents the basic concepts needed to understand this

research: Monitoring illegal drug trade and observing

the analysis of social media monitoring and social

media platforms from a point of view. Illegal drug

sales are illegal distribution of drug sin electronic

form (usually by coded speech or private electronic

messages). Natural Language Processing (NLP) is a

subclass of artificial intelligence (AI) that uses human

language to detect drug related rupes and lack.

 Use Cases Identify Objects Burgets (EG drugs)

in multimedia using image and video

recognition machine learning approaches

 Geographical Location Tracking: Using

location information from social media posts to

do track where drugs are being sold.

 Moral AI and Data Privacy: Regardless of the

moral guidelines and privacy requirements

when creating monitoring systems (such as

GDPR).

4 LITERATURE REVIEW

4.1 Key Concepts and Definitions

This section outlines fundamental concepts essential

to understanding the research:



Social Media Monitoring: Tracking and

analyzing con- tent on social platforms for

illegal drug sales.



Illicit Drug Sales: The illegal distribution of

narcotics through digital platforms, often

using coded language and direct messages.

•

Natural Language Processing (NLP): An AI

branch that analyzes human language, helping

detect suspicious conversations using drug-

related keywords and slang.

•

Image and Video Recognition: Machine

learning techniques for identifying objects, such

as drugs, in media posts.

•

Geolocation Tracking: Utilizing location data in

social media posts to identify drug sales hotspots.

•

Ethical AI and Data Privacy: Compliance with

ethical guidelines and privacy regulations, like

GDPR, in monitoring systems.

4.2 Historical Perspective

The role of social media in drug sales has evolved:

 Early 2000s: Platforms like My Space and

Facebooks aw minimal illegal activity.

 2010-2015: The rise of Instagram, Twitter, and

Snapchat led to increased drug trafficking due to

limited monitoring.

 2015-Present: Billions of users have made

social media lucrative marketplace for drugs,

complicating law enforcement efforts.

4.3 Theoretical Framework

Key frame works guiding this research include:

 Social Network Theory: Analyzes

communication pat- terns to detect criminal

behavior.

Investigating Drug Trafﬁcking Using Encrypted Messengers: NLP and Data Analysis Approaches in Cybersecurity

745

 Routine Activity Theory: Suggests crime

occurs when a motivated offender, suitable

target, and lack of guardian- ship converge.

 Machine Learning and Big Data Analytics:

Essential technologies for processing large

datasets to detect suspicious behavior.

4.4 Previous Research

Studies on social media monitoring, AI, and drug-

related crime detection include:

These studies emphasize the need for integrated

systems combining text analysis, image recognition,

and network analysis.

4.5 Reputation of the Area now

The topic of the subject is shifting rapidly, and here

are some key developments to keep an eye on:

•

Improvements in NLP: Pre-trained models

such as BERT and GPT are paving the way for

a better comprehension of illegal code.

Advanced drug recognition algorithms based in deep

learning algorithms, are achieving consistently

greater accuracy in detecting drugs on images from

social networks such as Instagram.

4.6 Identified Gaps

However, there are still critical gaps within the

monitoring systems:

Changing Language, decoding: Since drug dealers

often adapt their means of communication, they need

flexible systems to track it.

•

Infections: Most modern systems don’t offer

insight until after data has been processed post-

infection. We must establish systems that are

capable of collecting and processing

information in real time, which will make it

possible for police to react faster.

•

Multichannel Integration: There is a remarkable

lack of systems capable of pulling records from

different systems within minutes, hindering full

drug trafficking detection.

•

Ethics, Privacy and Security Concerns:

Developing ethical monitoring protocols would

involve balancing law enforcement needs with

individual rights.

Quickly addressing these gaps may help create better

tools for combatting drug sales through social media

and protecting the privacy rights of users. The table

shows the table 1: summary of authors, objectives,

and findings Filling these gaps will help build better

tools to combat drug sales on social media platforms,

while ensuring that ethical and privacy concerns are

not overlooked.

Table1: Summary of authors, objectives, and findings.

Authors Objective Findings

Huang et

al.

(2018)

Detects lang in

drug-related

conversations on

Twitter.

Defined and

identified

drugs lang in

tracking

illegal

transactions.

Smith et

al.(

2020)

Identification of

drug paraphernalia

in social media

images.

Develope

Accurate

system for

identifying

drug-related

items.

Garcia and

Flores(201

Use AI bots to

ascertain trends in

drug sales by

pretending to be a

buyer.

Successfully

gathered

insights into

drugs ales

trends using

ots.

Bakken

&Demant

(2019)

Study risk

perception by drug

vendors in social

media drug markets.

Public

platform

sellsers

perceived

higher risks

compared to

rivate ones.

Rhumor

barbeetal.

(2016)

Investigate Darknet

drug markets using

digital, physical,

and chemical data.

Found Dark

net markets

often offer

higher-quality

drugsat better

rices.

5 METHODOLOGY

5.1 Data Collection

The methodology was to expose the forensic data of

the experiment in encrypted chat logs and the dark

web market place. The algorithm is as follows:

•

Encrypted Chat Logs: Records of

experimental drug dealings were a hurdle in the

development of the data leakage.

•

Forensic Evidence: Cases of evidence were

identified from the impounded phones that were

legally seized from phones.

•

Dark Web Discussions: Chats on drug

trafficking were the topic of investigation.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

746

5.2 Digital Forensic Investigation

The inquiry followed the NIST methodology and was

bro- ken down into four major parts:

Collection: The most popular methods that

investigators use to collect evidence include the

MOBIL and Magnet Axiom forensic tools. These

tools are used to image WhatsApp messages and

other material, as well as analyze the files on SD cards

and SIM cards.

Examination: Each file is hashed and its integrity

measured. This ensures that the file cannot be

modified or tampered with, as any change in the hash

would alter the block, leading to loss of file integrity.

Analysis: Crime prevention models were used to

locate drug-related terms and collections of words

indicating a crime spree in favor of narcotics.

Reporting: The police crime labs and the scene

investigator must complete the remaining affidavit so

that the prosecution has the necessary evidence.

5.3 NTP Based Detection Model

The authors suggested a BERT based NLP model to

classify drug conversations from drug related

encrypted chat history. The model went like this:

 Preprocessing: Stop word and special character

removal from the messages, leading to

segmentation and noise elimination.

 Feature Extraction: The word embeddings

were used to understand the nuances and

meanings associated with the messages.

 Data Annotation and Reliability: If we want

the machine to have reliable data to learn from

then we also had to be careful with training data

so the model was trained on data obtained from

articles, TV Shows and Movies, etc, that had

relatable illegal activities

 Evaluation: The performance of the model was

evaluated using metrics like accuracy, precision

and recall.

5.4 Figures and Tables

Terms and groups of words suggesting a streak of

crimes favoring narcotics.

The figure 3 shows the Steps of NLP. And the

table 2 shows the Table 2 NLP based Detection

Model Stages. Finally, reporting: The police crime

labs and the scene investigator must fill out the

remaining affidavit so that the prosecution has what

they need.

Table 2: NLP based Detection Model Stages.

Stage Details

Preprocessing

Noise removal by deletings top

words and

Special characters.

Feature

Extraction

Word embeddings used to

capture relation

Ships and meanings of messages.

Model Training

Training on data from TV

shows, movies,

And articles featuring illegal

activities.

Evaluation

Performance measured through

accuracy,

precision, and recall.

Figure 3: Steps of NLP.

6 RESULTSANDDISCUSSION

6.1 Forensic Analysis of Encrypted

Message

Digital evidence such as deleted messages,

timestamps, and images was successfully extracted

from WhatsApp conversations using forensic tools.

The following key findings were observed:

A total of 67% of deleted messages were recovered,

demonstrating the effectiveness of forensic tools in

retrieving crucial evidence from suspects attempting

to erase their tracks.

100% of smartphone contacts were successfully

extracted, providing critical investigative leads for

law enforcement authorities.

Drug-related keywords were identified in 75% of

conversations, indicating that NLP-based analysis is

highly beneficial for detecting drugs muggling

activities through encrypted chats.

Investigating Drug Trafﬁcking Using Encrypted Messengers: NLP and Data Analysis Approaches in Cybersecurity

747

6.2 NLP Model Performance

The NLP model was based on BERT and achieved

the following performance metrics:

 Accuracy:91.2%

 Precision:89.5%

 Recall:87.8%

These results indicate that the NLP model is effective

in detecting drug-related discussions on encrypted

messaging platforms. The combination of high

precision and recall ensures reliability by minimizing

false positives while capturing relevant crime-related

messages.

6.3 Implications for Law Enforcement

The integration of NLP-based monitoring with digital

forensic analysis provides significant advantages for

law enforcement agencies:

 Detection of Illicit Transactions: Secure

internet communications often create challenges

for investigations, but NLP-assisted monitoring

enables law enforcement to detect and track

potential drug transactions.

 Automated Text Analysis: NLP algorithms

facilitate the rapid processing of large volumes

of text data, allowing for quicker investigations

and real-time responses to criminal activities.

 Recovery of Critical Evidence: Digital

forensic tools, when combined with text

analysis techniques, can extract keywords and

topics from deleted messages and multimedia

files, providing crucial evidence admissible in

court.

6.4 Application of the Methodology: A

Case Example

The methodology was applied to data collected from

Insta- gram posts and comments to detect and analyze

potential drug trafficking activities. The workflow

illustrates the system’s capability to uncover hidden

patterns and relationships within the data. Below is

the detailed example highlighting each step:

Data Collection and Storage: Instagram posts,

comments, and hashtags relevant to drug-related

activities were gathered. Specific hashtags such as

#acidtrip and #lsdtabs formed the thematic focus of

the dataset. A total of 12,857 posts were securely

stored for further processing to maintain data integrity

and enable advanced computational analysis.

Comments and Hashtags Analysis: The following

combinations of comments and hashtags were

identified:

CommentC1: Included #acidtrip and #lsdtabs,

indicating potential connections.

CommentC2: Focused so lelyon #acidtrip.

Additional comments: Showed varied and recurring

combinations of related hashtags.

These findings form the basis for uncovering patterns

within the data.

Graph Representation: To visualize the

relationships between hashtags, a graph structure

was created:

Nodes represented hash tags, suchasH1(#lsdtabs)

and

H3(#acidtrip).

Edges depicted Connections between nodes. For

instance, #lsd tabs (H1) was linked to #acidtrip (H3),

and #acid (H2) was also associated with #acidtrip

(H3).

This graph served as a critical tool to understand the

underlying network of hashtags.

Matrix Representation: The graph relationships were

converted into a matrix format, enabling

computational analysis:

Rows and columns represented in dividual hashtags.

Matrix values indicated the strength or presence of

links between pairs of hashtags.

This representation facilitated further processing with

advanced mathematical techniques.

Graph-Based Deep Learning Implementation:

A graph based deep learning model was employed to

analyze the data:

 Integrated graph and matrix representations

provided in- sights into hidden patterns.

 The model demonstrated the ability to detect

key trends, connections, and clusters within the

data.

 For instance, the analysis revealed 1,228 flagged

posts and 267 distinct user accounts associated

with potential drug-related activities.

Figure 4: Example.

7 DISCUSSION

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

748

The findings of this case study highlight the

effectiveness of the proposed method for finding drug

related activities on social media. Graph-based DEEP

Wanda with traditional data analysis techniques can

consolidate education models, system:

Identify and test their current pattern in hash tags and

user comments. Provide a valuable understanding of

human trafficking network, law enforcement efforts.

Adapt to the development language and coded

terminology used in illegal activities.

This case study emphasizes the possibility of

connecting NLP and graph-based models for

experimental cybersecurity applications, especially

fighting the DRUG trafficking.

This example highlights the potential of integrating

NLP and graph-based models for real-world

applications in cybersecurity.

8 CONCLUSIONS AND

FUTUREWORK

This look at emphasizes the potential of NLP mixed

with digital forensics in detecting and preventing drug

trafficking on encrypted messaging systems. The

consequences display that at the same time as

machine getting to know models can discover illegal

activities, forensic tools play a critical role in getting

better crucial proof that criminals attempt to erase.

Future research should focus on:

Real-Time Detection: Developing deep learning

models that can recognize illegal activities in real-

time.

Forensic Analysis of Other Platforms: Expanding

the forensic framework to other encrypted messaging

applications, such as Signal.

Legal Frameworks: Establishing legal systems that

balance the need for encrypted communication

monitoring while protecting user privacy.

The integration of AI and digital forensics in law

enforcement will enhance authority’s ability to

predict drug traffickers’ operational models and

contribute to as after digital world.

REFERENCES

Aldridge, J., &De´cary-He´tu, D.,” Hidden wholesale: The

drug diffusing capacity of online drug cryp to markets,”

The International Journal on Drug Policy, vol. 35, pp.

7–15, 2016.

Bacon, M.,” Desistance from criminal is a tion: police

culture and new directions in drugs policing,” Policing

and Society, vol. 32, no. 4, pp.522–539, 2022.

Bakken & Demant,” Sellers’ risk perceptions in public and

private social media drug markets,” 2019. Objective:

Risk perception by drug vendors in social media drug

markets. Findings: Sellers on public platforms per-

ceived higher risks than private platforms.

Bakken, S. A.,” App-based textual interviews: interacting

with younger generations in a digitalized social reality,”

International Journal ofSocial Research Methodology,

vol. 0, no. 0, pp. 1–14, 2022.

Cabrera-Nguyen,E.P.,Cavazos-

Rehg,P.,Krauss,M.,Bierut,L.J.,& Moreno, M. A.,”

Young adults’ exposure to alcohol-and marijuana-

related content on Twitter,” Journal of Studies on Al-

cohol and Drugs, vol. 77, no. 2, pp. 349–353, 2016.

Chuanbo, Hu., Bing, Liu., Yanfang, Ye., & Xin, Li.,” Fine-

grained Classification of Drug Trafficking Based on

Instagram Hashtags,” Decision Support Systems, 2022.

DOI: 10.1016/j.dss.2022.113896.

Chuanbo, Hu., Bing, Liu., Xin, Li., &Yanfang, Ye.,” Un-

veiling the Potential of Knowledge-Prompted ChatGPT

for Enhancing Drug Trafficking Detection

on Social Media,” arXiv.org, 2023.

DOI:10.48550/arXiv.2307.03699.

Garcia and Flores,” AI-led bots to interact with drug deal-

ers,” 2019.Objective: Utilize AI bots to ascertain trends

in drug sales by pretending to be a buyer. Findings:

Successfully accrued drug sales trends insights using

bots.

Huang et al.,” NLP for Twitter drug-related conversation

detection,”2018. Objective: Detect slang in drug-

related conversations on Twitter. Findings: Defined and

identified drug slang in tracking illegal transactions.

Kang, R., Dabbish, L., Fruchter, N., &Kiesler, S.,” “My

Data Just Goes Everywhere:” User Mental Mod-

elsoftheInternetandImplicationsfor Privacy and Securi-

ty,” Eleventh Symposium on Usable Privacy and Secu-

rity (SOUPS 2015), Ottawa, Canada, 2015.

M. J.,” Predictors of buying drugs on social media among

young people in New Zealand: Findings from a large

online survey,” International Journal of Drug Policy,

vol. 98, p. 103430, 2021.

Moreno, M. A., & Whitehill, J. M.,” #Wasted: The inter-

section of substance use behaviors and social media in

adolescents and young adults,” Current Opinion in

Psychology, vol. 9, pp. 72–76, 2016.

Moyle, L., Childs, A., Coomber, R., & Barratt, M. J.,

”#Drugsforsale:Anexplorationoftheuseofsocialmediaa

ndencryptedmessagingappsto supply and access

drugs,” The International Journal on Drug Policy, vol.

63, pp. 101–110, 2019.

Rhumorbarbe et al.,” Buying drugs on a Darknet market: A

better deal?” 2016. Objective: Investigate Darknet drug

markets using digital, physical, and chemical data.

Findings: Found higher-quality drugs at better prices on

Darknet markets.

Ryan, J. E., Smeltzer, S. C., &Sharts-Hopko, N. C.,” Chal-

lenges to studying illicit drug users,” Journal of Nursing

Scholarship, vol. 51, no.4, pp. 480–488, 2019.

Shah, N., Li, J., & Mackey, T. K.,” An unsupervised ma-

chine learning approach for the detection and charac-

Investigating Drug Trafﬁcking Using Encrypted Messengers: NLP and Data Analysis Approaches in Cybersecurity

749

terization of illicit drug-dealing comments and interac-

tions on Instagram,” Substance Abuse, vol.43, no.1, pp.

273–277, 2022.

Smith et al.,” Computer vision for the detection of Insta-

gram and Snapchat drug paraphernalia,” 2020. Objec-

tive: Identification of drug paraphernalia in social me-

dia images. Findings: Developed an accurate system to

identify drug-related items.

Tianyi Ma, Yiyue Qian, Chuxu Zhang, &Yanfang Ye,”

HypergraphContrastiveLearningforDrugTraffick-

ingCommunityDetection,”DOI:10.1109/icdm58522.20

23.00149, 2023.

Trottier, D., Social Media as Surveillance: Rethinking

visibility in a converging world, Routledge, 2012.

Vander Sanden,R., Wilkins ,C .,Romeo ,J.S.,Rychert,M.,

&Barratt,

Vannucci, A., Ohannessian, C. M., & Gagnon, S.,” Multi-

site social media use and psychological adjustment

during emerging adulthood,” Emerging Adulthood, vol.

7, no. 6, pp. 501–506, 2019.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

750