Threatening language detection from Urdu data with deep sequential model.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Social Media Analysis

Background:

• Urdu language processing (ULP) faces challenges on social media due to a lack of specialized libraries.
• Existing methods for Urdu data analysis are hindered by the absence of comprehensive online Urdu vocabularies.
• Efficient preprocessing of Urdu text is crucial for accurate threat identification on platforms like Twitter and Facebook.

Purpose of the Study:

• To develop essential Urdu Language Processing (ULP) resources, including a stop words list and stemming dictionary.
• To enable efficient data preprocessing for Urdu text, comparable to English language standards.
• To enhance the accuracy of threat detection in Urdu social media data using advanced machine learning models.

Main Methods:

• Creation of an Urdu language vocabulary, incorporating a stop words list and a stemming dictionary.
• Implementation of data cleaning and stemming techniques to reduce input size and remove noise from Urdu sentences.
• Training and evaluation of a deep sequential model utilizing Long Short-Term Memory (LSTM) units on the preprocessed Urdu data.

Main Results:

• The developed Urdu preprocessing tools effectively reduced sentence input size and removed redundant information.
• The Long Short-Term Memory (LSTM) model achieved a prediction accuracy of 82% on the processed Urdu social media data.
• The proposed methodology demonstrated superior performance compared to existing Urdu language processing and threat detection methods.

Conclusions:

• The introduction of Urdu-specific preprocessing resources significantly improves the efficiency of Urdu Language Processing (ULP).
• The developed LSTM-based model shows high potential for accurate threat detection in Urdu social media content.
• This research provides a foundational step towards more robust Urdu natural language understanding and analysis.