Narcotrace: Advanced Detection System for Social Media-Based

Drug Trafficking

Khalid Alfatmi, Aarya Borse, Rushikesh Girase, Sakshi Bagul, Abhijit Patil and Makarand Shahade

Department of Computer Engineering SVKM’s Institute of Technology Dhule, India

Keywords: Drug Trafficking Detection, Machine Learning, Real-Time Threat Assessment, Social Media Analysis, Risk

Categorization.

Abstract: Platforms like Telegram, WhatsApp, and Instagram have become very intelligent and hard to follow and track

for law enforcement agencies in terms of the oversight and detection of criminal activities being done across

borders of drug trafficking. Based on this, this paper introduces NarcoTrace as an innovative software

application intended for countering real-time occurrences of drug trafficking on social media. NarcoTrace

will utilize third-party APIs along with advanced models of machine learning that would include BERT in

text analysis to analyze massive datasets of user-generated content while offering actionable insights.

NarcoTrace incorporates an assessment system that ranks and categorizes threat subjects from high, medium,

to low risk using immediate analysis of data. Activity types, ranked as medium risk are colored Yellow while

high-risk activity types are colored Red. Hence, law enforcement shall concentrate on the most threatening

areas more quickly. The usability of the dashboard is enhanced by incorporating metadata that contains the

user IDs, timestamps and logs of the activities. This paper details the architecture of the system, methodologies

data. Finally, we discuss potential societal impacts and future directions, which include an expansion of

NarcoTrace to recognize emerging trends in criminal activity.

1 INTRODUCTION

Communication has really improved regarding the

use of social media, whereby through instant

connections one has been reached across the globe.

However, such a digital revolution brings fertile

grounds for criminal activities, especially drug

trafficking. Thus, in this regard a platform such as

Telegram (chatbots) or WhatsApp would offer

protection for secrecy of user privacy while keeping

illicit operations undetected. These drug traffickers

use such media platforms for penetrating into larger

audiences, arranging transactions, and evading law

enforcement. Thus, to confront them in such

complications, advanced technological solutions are

required. Unlike the traditional approach to

combating drug trafficking, which involves human

observation, gathering of intelligence data is

extremely time-consuming and susceptible to human

error. The encryption used in modern communication

tends to limit the effectiveness of the conventional

surveillance tools. In order to attain better accuracy in

the identification of criminal activities. Therefore,

automated systems that have the ability to analyze

large amounts of data in real-time have been

challenged. NarcoTrace addresses this challenge with

the use of third-party APIs and complex machine

learning algorithms that can scan textual data in

detail. It uses BERT for its analysis of text to identify

linguistic flavor pertaining to drug-related activities.

A strong backend supports it, where it also gathers

metadata, including IP addresses, email IDs and

timestamps for info crucial to the investigative

agencies.

NarcoTrace excels in being able to perform live

risk assessment of the activities of the user into

different levels of risk. This enables law enforcement

agencies to pinpoint which cases are at high risk and

make sound resource allocation decisions and deploy

forces accordingly. Reports and alert information

related to any critical data are simply thrown on the

dashboard of the system’s user interface. NarcoTrace

is the next giant leap in cybersecurity and law

enforcement. It can scale to change and grow into

new platforms and unpredictable threats. This paper

focuses on the design, implementation and

Alfatmi, K., Borse, A., Girase, R., Bagul, S., Patil, A. and Shahade, M.

Narcotrace: Advanced Detection System for Social Media-Based Drug Trafﬁcking.

DOI: 10.5220/0013659200004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 757-762

ISBN: 978-989-758-763-4

757

performance of NarcoTrace and brings out its

potential to revolutionize the digital war against drug

trafficking

2 LITERATURE SURVEY

2.1 Overview of Current Method of

Detection

Drug trafficking circles have proved to be very elastic

when it comes to applying social media apps like

Telegram, WhatsApp and Instagram as messaging

instruments. All of these offer anonymity, a very

large user base that lets traffickers hide in them and

behave quite undetected. Trafficking scales as well as

techniques used make old detection mechanisms, for

example manual content moderation and keyword

searching ineffective (Alves and Pedroso, 2019),

(Ma, Qian, et al. , 2024). Recent studies have,

therefore, started to search for using machine learning

and AI in this matter. For instance, Alves et al.

employed the use of hypergraph learning method on

drug trafficking networks in the pursuit of gaining

insight in the community structures on which

traffickers operate and converge on social media

platform 3. Similarly, Shah et al. employed the social

network analysis on the encrypted communication

where evidence was realised that at the network level,

it can be a point for the detection of illicit activities 4.

However, they lack to hold the strength to tackle the

comprehensive use of multimodality in the content of

the social media relating to the usage of text, images

and audio.

2.2 Text Analysis in Drug Trafficking

Detection

It remains however one of the most prominent types

of data used for the passing on of information among

traffickers. It majorly relied on shallow text mining

approaches: TF-IDF and keyword matching, in

earlier approaches. These do not understand the

contexts and are not up to nuanced language analysis

based on context understanding (Alves, 2023),

(Wang, 2021). Ma et al have developed sophisticated

NLP models, BERT, to detect the languages related

to drugs on social media. These models are better at

subtlety in linguistic patterns and context than the

traditional method approach (Shah and Sen, 2020).

Even though the NLP models are highly effective,

there is still a huge challenge in using the constantly

evolving slang used in dealing with drug trafficking.

In fact, baselines such as LDA for topic modeling are

merged into bigger frameworks that almost captured

the semantic richness of user conversations (Desai

and Patel, 2019). The NarcoTrace models used the

concept of transfer learning wherein finetuned pre-

trained models can learn through specific data sets to

enhance its ability in adapting new slang and jargon

(Shah and Sen, 2020).

2.3 Image and Video Analysis

Images and videos are another form of visual

information which are commonly used by traffickers

for advertising their products or confirming

transactions. Object detection models based on

YOLO have been promising in identifying drug

paraphernalia within images. Alves et al. underlined

the employment of YOLO V5 to identify secret

shipments of drugs, emphasizing its real-time

detection capabilities with high accuracy (Shah and

Sen, 2020), (Kumar and Gupta, 2020). However,

traffickers usually use obfuscation attacks, like

altering a view of an object or hiding prohibited

objects in seemingly benign events. That is

particularly challenging even for as advanced models

as YOLO. In response to this, NarcoTrace uses

advanced preprocessing techniques to increase its

robustness against such evasion attacks (Kumar and

Gupta, 2020).

2.4 Audio Analysis for Crime Detection

Audio communications are favored by traffickers

since they are seen as untraceable. Nowadays, with

speech recognition models including Wav2vec,

transcript and analysis of audio messages might now

potentially be made possible (Shah, 2023). Honourio

et al. even demoed how audio analysis may aid

forensic investigations by picking crucial phrases

indicating drug trafficking activities.

Despite these developments, there are still issues

with how it deals with mixed accents, noise, and

audio compression. NarcoTrace utilizes noise

reduction and sophisticated models of speech to make

sure its detection capability is insensitive to the

adverse effects of poor audio quality conditions ,

(Shah, 2023).

2.5 Limitations and Gaps

Although some notable progresses have been made,

outstanding gaps still persist in detecting online drug

trafficking via social media. Current systems lack the

processing of several data modalities simultaneously

INCOFT 2025 - International Conference on Futuristic Technology

758

and, therefore, realize fragmented insights and lower

detection accuracy (Shah and Sen, 2020), (Kumar and

Gupta, 2020). Furthermore, the large amount of

labeled datasets has a reliance on adapting much

when changing with regards to the type of illicit

activity for training the model. Narcotrace fills in the

gap by combining text, image, and audio analysis

under a single framework. Ensemble learning

improves the robustness between data types and self-

supervised learning minimizes reliance on labeled

data, thereby adapting minimal retraining to new

emerging threats (Shah and Sen, 2020), (Kumar and

Gupta, 2020) , (Shah, 2023).

3 PROPOSED SYSTEM

3.1 System Overview

The system to be proposed will be a software program

created specifically for the detection and

identification of activities on Telegram related to drug

trafficking. This shall be the basis for the approach

wherein multimodal analysis will integrate text,

image and audio analyses together into one

comprehensive detecting and monitoring system. For

these bases, advanced models such as YOLO V5 for

object detection in pictures, Wav2vec for audio

analysis and BERT for text analysis can be used. This

would allow the detection of illegal activities and

material in Telegram chats and channels as

effectively as possible.

3.1.1 Architectural Overview: The overall

architecture of NarcoTrace is

segmented into three modules

3.1.1.1 Data Collection Module

Using TDLib API, it collects all activities and

metadata conducted by users within the Telegram

system.

3.1.1.2 Analysis Module

A set of machine learning models is used to process

and analyze the collected data.

3.1.1.3 Dashboard User Interface

NarcoTrace empowers law enforcement agencies to

take immediate action through real-time

visualization, risk assessment and alerting.

3.1.2 Risk Assessment and Alerts:

Based on the analytical results, NarcoTrace profiles

Telegram users through a risk assessment framework:

Low Risk (Green): Less evidence of illegal activities

Medium Risk (Yellow): Activity is present but raises

suspicion. High Risk (Red): Concrete evidence of

drug trafficking or other activities which require

attention.

Figure 1: Architecture Diagram

3.1.3 System Deployment and Scalability

The NarcoTrace system will be deployed within a

cloud environment, which ensures scalability and the

reliability to handle a large volume of Telegram data.

It also means that it can easily be updated, accounting

for the shifts and change in needs and user behavior.

3.1.4 Dashboard Interface

Dashboard interface puts emphasis on usability and

functionality:

a. Real-Time Notifications: Color-coded

notifications depict risk levels.

b. Activity Logs: Activity logs which list

interactions on the part of the user,

including message content and timestamps.

Narcotrace: Advanced Detection System for Social Media-Based Drug Trafﬁcking

759

c. Metadata Visualization: Graphical

representations of user activity trends and

patterns.

3.1.5 Privacy and Ethics

NarcoTrace is very sensitive when it comes to data. It

therefore adheres to strict privacy and ethical

guidelines. All stages of the transmission process and

storage involve encryption used in the application to

ensure data safety. The legal frameworks guiding

ethical use are adhered to within the jurisdiction of its

operation.

Figure 2: Dashboard Design

3.2 Methodology

The methodology put in place by the NarcoTrace

system primarily consists of real-time monitoring,

machine learning-based analytical approaches, and

comprehensive data collection mechanisms that aim

at effectively detecting drug trafficking activities

through Telegram. This paper develops procedures

for gathering data, trains the dataset specialized in

machine learning models, and produces real-time

alerts and insights for law enforcement regarding the

illicit trades through Telegram.

3.2.1 Data Collection and Preprocessing:

The Data Collection Module acts as the backbone for

real-time monitoring by retrieving metadata from the

Telegram API. It fetches user IDs, timestamps, IP

addresses, and other content including text, images,

and audio. After data collection, it gets preprocessed

for quality and relevance in the following manners:

a. Text Preprocessing: Irrelevant information,

special characters, and stop words are

removed from Telegram messages.

b. Image Scaling and Filtering: Images are

filtered and sharpened to make them ready

for object detection.

c. Audio Segmentation: A large number of

long audio files have been broken down to

reasonable sizes allowing for further

analysis.

3.2.2 Machine Learning Models:

Three Optimized models are the crux of machine

learning analytics of NarcoTrace-the three models

optimized for using the three types of data on

Telegram.

a. Text Classification with BERT: It employs

BERT, Bidirectional Encoder

Representations from Transformers, to

analyze text on Telegram. This model

actually does well in detecting contextual

language, including encrypted slangs for

drugs. Fine-tuning with domain-specific

datasets would improve it in identifying

suspicious communications.

b. Image Analysis (YOLO V5): This is the

object detection algorithm used for real-time

object detection in images coming from

Telegram related to narcotics trafficking,

including the packaging material that may

be related to those distributions.

c. Audio Analysis (Wav2vec): The Wav2vec

analyzes audio file transcription and analysis

with the key concepts and related speech

phrases about drug trafficking, even in noisy

environments and diverse speech accents.

Figure 3: Results

INCOFT 2025 - International Conference on Futuristic Technology

760

4 LIMITATIONS AND FUTURE

WORK

With all this notwithstanding, however, several

challenges and limitations remain with NarcoTrace.

The technique powers tracking and detection of drug

trafficking through Telegram but leaves more

challenges behind.

4.1 Limitations

Even though NarcoTrace is as potent as it can get,

many of its drawbacks impact the overall efficiency

with which the mission is furthered.

a. Evasion Techniques: NarcoTrace presented

the fact that even though their system was

detecting a certain strategy being employed

by drug dealers, the dealers kept changing.

This indicates that their changes should be

input into the system, hence having patterns

and tactics that ought to be well understood

by the system.

b. Dependence on Telegram APIs: NarcoTrace

is based on the available API from

Telegram. Therefore, the performance of

NarcoTrace would depend on the imposed

restrictions or limitations on the access to the

API.

c. Accuracy of Models: With such a few false

positives and negatives in NarcoTrace, there

are many places that call for fine-tuning

models and datasets, where improvement is

looked into in detecting errors as well as

improved precision.

4.2 Future Improvements:

Overcoming Challenges and Strengthening

NarcoTrace Future versions of the system would

improve on the following to make the system more

robust.

a. Improved Contextual Understanding:

Refine the core ingredient of natural

language processing towards enhanced

contextual recognition in a conversation,

thus lowering false positives and further

increasing the rate of identification

accuracy.

b. Obfuscation Detection: Improve its

detection capacity by having a special focus

on finding hidden and obfuscated content

such as masked text, as well as encrypted

communication from the traffickers on

Telegram.

c. Developing new functionalities to polish the

mechanism of detection against the trend of

evolution and usage on Telegram.

d. Automated Compliance Monitoring: That

such automation be used both for monitoring

and data handling in respect of laws of

innate privacy and the proper regulatory

guidelines which could provide a safe,

privacy-protected environment for users and

an even more trusted environment for the

public at large.

5 CONCLUSIONS

The advent of social media channels such as

Telegram, WhatsApp, and Instagram has changed the

way information is exchanged with people, instantly

and in the most expansive ways possible. In the

process, the same digital progress gives illegal

activities, especially drug trafficking, an easy life

under the cover of encryption and private messaging.

In this direction of real-time identification and

tracking of events associated with drug-related

activities over platforms like Telegram, NarcoTrace

emerges as a leading-edge solution. NarcoTrace uses

third-party APIs and advanced machine learning,

especially natural language processing models like

BERT, to analyze the textual data involved with

illegal activity. While the system has plans for future

capabilities regarding content analysis of images and

audio, at present, the text analysis is the main focus,

and such can enable it to come up with a more

effective combination in identifying suspicious

interactions. Highly scalable and adaptable toward

adapting big volumes of data within a cloud-based

infrastructure without disturbing the operational

process on dynamic platforms like Telegram. The

most relevant aspect is that NarcoTrace has an inbuilt,

real-time risk assessment mechanism that categorises

all the users into low, medium, and high-risk ranks

based on behaviour patterns. As such, it would be

made possible for law enforcement to intervene fast

but where only necessary and without compromising

accuracy where false positives and negatives have

been minimized to a great extent and with the system

remaining aligned to the legal and ethical standards

regarding privacy issues, thereby lending much

credence to public trust in such systems. It is a big

stride in the battle against drug trafficking on social

media by equipping law enforcement with state-of-

the-art tools of detection and prevention. NarcoTrace

protects vulnerable populations from contact

exposure but also decreases broader societal and

Narcotrace: Advanced Detection System for Social Media-Based Drug Trafﬁcking

761

economic costs of drug abuse through improving

operational precision. With these features,

NarcoTrace has the potential to enhance public

safety; the frontline of defense against narcotics

trafficking is to be increased by putting into place

secure data policies to protect the rights and privacy

of its users.

REFERENCES

Alves, J., and Pedroso, H. A. C. G., ”Detecting Relevant

Information in High-Volume Chat Logs: Keyphrase

Extraction for Grooming and Drug Dealing Forensic

Analysis,” in Proceedings of the IEEE International

Conference on Data Mining (ICDM), 2019.

Ma, T., Qian, Y., Zhang, C., and Ye, Y., ”Hypergraph

Contrastive Learning for Drug Trafficking Community

Detection,” in IEEE International Conference on Data

Mining (ICDM), 2023.

Shah, K., and Sen, A., ”Monitoring Individuals in Drug

Trafficking Organizations: A Social Network

Analysis,” in Proceedings of the IEEE/ACM

International Conference on Advances in Social

Networks Analysis and Mining (ASONAM), 2020.

Zhang, L., Wang, Y., and Li, F., ”Machine Learning in

Food Safety: A Case Study on Milk Adulteration

Detection,” IEEE Transactions on Automation Science

and Engineering, vol. 16, no. 3, pp. 1072-1080, July

2019.

Honorio Alves, J., ”Detecting Drug Trafficking Activities

through Deep Neural Networks in Social Media

Contexts,” in Proceedings of ICMLA 2023.

Wang, T., ”Distilling Meta-Knowledge on Heterogeneous

Graphs for Illicit Drug Trafficker Detection,” in

Advances in Neural Information Processing Systems

(NeurIPS), 2021.

Desai, M., and Patel, T., ”IoT-Based Real-Time Monitoring

of Illicit Activities on Encrypted Platforms,” in

International Research Journal of Engineering and

Technology (IRJET), vol. 6, no. 11, pp. 159-166, Nov.

2019.

Kumar, S., and Gupta, P., ”A Survey on AI and

Cybersecurity in Drug Trafficking Detection,” Applied

Sciences, vol. 10, no. 9, pp. 5271, 2020.

Shah, P., ”AI for Social Media Monitoring: Challenges and

Opportunities in Encrypted Environments,” Journal of

Internet Research, 2023.

INCOFT 2025 - International Conference on Futuristic Technology

762