MARSAD: A Multi-Functional Tool for Real-Time Social Media Analysis

Md. Rafiul Biswas, Firoj Alam, Wajdi Zaghouani · 2025

Read MARSAD: A Multi-Functional Tool for Real-Time Social Media Analysis by Md. Rafiul Biswas, Firoj Alam, Wajdi Zaghouani (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-01369.txt.

Introduction

Rafiul Biswas, Firoj Alam, and Wajdi Zaghouani built a tool for live social media study. The tool, MARSAD, focuses on the Arabic-speaking world. It lets users with no code skill study live or saved posts. Clear charts and reports come out with a few clicks.

The tool checks mood, feeling, and hate speech in posts. It also flags propaganda and checks claims for truth. Safe scraping with API keys pulls in public posts. Such a tool helps news staff, scholars, and groups track fast shifts in online talk across the region.

Catalog record

CREATORMd. Rafiul Biswas, Firoj Alam, Wajdi Zaghouani
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-01369

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MARSAD: A Multi-Functional Tool for Real-Time Social Media Analysis Md. Rafiul Biswas, Firoj Alam, Wajdi Zaghouani Abstract MARSAD is a multi-functional natural language processing (NLP) tool designed for real-time social media monitoring and analysis, with a focus on the Arabic-speaking world. It empowers researchers and users without coding expertise to analyze both live and archived social media data, generating detailed visualizations and reports across dimensions such as sentiment analysis, emotion analysis, propaganda detection, fact-checking, and hate speech detection. MARSAD also offers secure data scraping capabilities using API keys for public social media data. MARSAD’s backend architecture combines flexible document storage with structured data management, enabling efficient handling of large and multimodal datasets. The frontend is designed to be user-friendly, allowing for seamless data upload and interaction. This paper outlines the architecture, key features, and future enhancements of MARSAD, positioning it as an essential tool for social media analytics. 1 1 1 A video demonstrating the system can be found at https://youtu.be/Bln8iSvwtWg?si=fLSfI4cpmbNSks33. 1 Introduction MARSAD aims to empower users—ranging from researchers to media analysts—with insights into public opinion and social media dynamics. By providing in-depth analysis of online conversations, the tool helps identify influential figures, track trending topics, and find the sentiment of discussions. This tool leverages advanced Natural Language Processing (NLP) technologies and traditional software development technologies. There are some existing tools like Communalytic 2 2 2 https://communalytic.org/ that allow users to download data and perform different analyses such as sentiment, toxicity, and topic analyzer and others. They use text embedding from pre-trained transformer-based models to find topics dynamically without human involvement. The core of the process involves converting the text (social media posts, in this case) into embeddings, which are numerical representations of the text’s meaning. Commonly, pretrained transformer-based models such as BERT 5 , RoBERTa 10 , Sentence-BERT (SBERT) 13 , or word2vec are used to generate these embeddings. Then, they classify the text by using clustering and labeling them. However, this application mainly focuses on English, French, and German languages. European News brief EMN News Brief 3 3 3 https://emm.newsbrief.eu accumulated news sources from online news portals and provides a summary of results that got the highest attraction among news sources. However, they focus only on news channels and do not focus on social media channels. Another tool, social media content Repear 4 4 4 https://reaper.social/ , scrapes data from social media. By providing access tokens, API keys, and search parameters, this tool enables downloading public content available on social media. NLP tool Social Media Sentiment Visualization 5 5 5 https://www.csc2.ncsu.edu/faculty/healey/social-media-viz/production/ can perform real-time analysis, but it is limited to Reddit and Threads. Also, it does not allow the uploading of user files and analysis of the data. Tanbih 6 6 6 https://tanbih.qcri.org/ allows users to identify propagandist content 15 , fact-checking 16 , bias detection 14 , and clustering, especially in the Arabic language article. What sets MARSAD apart is its exclusive focus on supporting the Arabic language, and it integrates the significant types of analysis (e.g., sentiment, propaganda, fact-checking) considerably in a single platform 3 ; 2 . MARSAD is capable of handling multiple user requests and performing several tasks simultaneously. Content that does not require an API key, such as Reddit and online news portals, can be analyzed in real-time and provide different kinds of visualizations and summary results. By providing the API key and access token obtained from social media (e.g., Facebook, Twitter, YouTube), MARSAD will work as a data crawling platform, same as CrowdTangle. 7 7 7 https://transparency.meta.com/researchtools/other-datasets/crowdtangle Moreover, it includes the integration of five major regional Arabic dialects, making it the first social media monitoring tool to offer such comprehensive linguistic coverage in Arabic. Through its state-of-the-art NLP tools, MARSAD handles the linguistic diversity of the region, accommodating variations in dialects and enabling more accurate and culturally relevant analysis. Additionally, we plan to integrate Arabic-named entity recognition 12 , enabling the identification of proper names, geographical locations, and organizations within social media posts 11 . 2 MARSAD Architecture MARSAD architectural system can be divided into two major part: i) frontend part ii) backend parts. Figure 1 depicts the overview of the system architecture. Frontend : In the frontend, the system interacts directly with the end users, who upload datasets that can include text, images, or multimodal data (combinations of text and images). The system is designed to accept various data formats, such as CSV, TSV, and JSON/JSONL. Once the data is submitted, it is validated with the structure and integrity of the uploaded data. It checks for missing values, correct field types, and schema compliance. It is performed by the Data Handler framework, by the Data Handler framework, which has been developed using modern JavaScript frameworks and libraries, including Next.js for server-side rendering, Flux for unidirectional data flow, and Node.js for server-side operations. After the validation process, the system generates metadata from the input data and stores it in the appropriate databases. The figure shows this flow, where user-provided data moves through the frontend, is processed by the Data Handler, and then transferred to the backend for further operations. Once the data is successfully processed and stored, it is made available for visualization. The visualization component allows users to explore and analyze the results through a user-friendly interface, likely including graphical representations of the analyzed data. Backend : In the backend, the system manages the complex processing, analysis, and storage tasks, ensuring efficient handling of large volumes of data. The data is stored in a hybrid database architecture utilizing both MongoDB 4 for flexible, schema-less document storage and PostgreSQL for structured, relational data management. This combination allows the system to scale effectively while handling diverse datasets, including multimodal, structured, and unstructured data. A core component of the backend is the Model API , which serves as the interface between the application and the underlying NLP models. This API is responsible for coordinating requests for various machine-learning models, including sentiment analysis, propaganda detection, and other NLP tasks. These models are hosted in a containerized environment (e.g., using Docker) to ensure scalability, isolation, and seamless deployment of different versions 6 . The Model API leverages modern deep learning frameworks such as TensorFlow, PyTorch, and Transformers, enabling it to load, fine-tune, and infer from pre-trained models like AraBERT, MARERT 1 , CamelBERT 8 , or custom models for specialized tasks like propaganda detection. It supports asynchronous processing to handle high-throughput requests and can manage multiple models simultaneously, ensuring low-latency responses and robust performance even under heavy load. Additionally, the backend side includes task queues (e.g., Celery) 8 8 8 https://github.com/celery/celery to manage long-running or resource-intensive operations like model inference or dataset processing. These queues allow the system to prioritize tasks and manage concurrency efficiently, enabling real-time user interactions with the frontend while delegating complex analyses to the background. Security and data integrity are maintained through the use of authentication and authorization mechanisms (e.g., OAuth) to ensure that only authenticated users can access and manipulate data or trigger model executions. Data is encrypted both at rest and in transit using AES encryption and SSL/TLS protocols, ensuring compliance with data privacy standards. After the analysis is completed, the results are passed back through the system, where

Read the complete text