README.md

metadata

language:
  - en
pipeline_tag: text-classification
license: agpl-3.0
tags:
  - not-for-all-audiences

LUNA-ProACT - PROtective Analysis for Cyber Threats

LUNA-ProACT is a context-aware machine learning model designed to protect users from online threats, primarily focusing on cybergrooming. It is created to work in association with LUNA, an AI-powered cybergrooming prevention application.

LUNA-ProACT uses natural language processing (NLP) to analyze language use in real-time and detects unusual patterns that could indicate potential online threats. It operates locally on iOS and Android mobile devices using TensorFlow Lite, providing real-time protection without requiring network traffic.

This model results from a collective effort during the TRUE CRIME HACKATHON 2023. Special thanks go to the Polizei Niedersachsen for inspiring the LUNA Cybergrooming Prevention App idea.

Key Features

• Real-Time Analysis: LUNA-ProACT scans text input on the fly, examining for patterns and language use typically associated with cyber threats.
• Context Awareness: Recognizes cultural, geographical, and age-specific contexts for accurate and relevant threat detection.
• Local Processing: Employs TensorFlow Lite to analyze data directly on the user's device, ensuring maximum privacy.

Dependencies

• Python 3.7+
• TensorFlow 2.5.0
• HuggingFace Transformers

Usage

1. Installation: Import the LUNA-ProACT model into your Python project.

   from transformers import AutoTokenizer, TFAutoModel
   tokenizer = AutoTokenizer.from_pretrained("LUNA-ProACT")
   model = TFAutoModel.from_pretrained("LUNA-ProACT")
   

2. Inference: Use the model for threat detection.

   input_text = "Your text here..."
   inputs = tokenizer.encode(input_text, return_tensors='tf')
   outputs = model(inputs)[0]
   
   # Outputs return the likelihood of the text being a potential cyber threat.
   threat_probability = tf.sigmoid(outputs)
   
   if threat_probability > 0.5:
       print("Potential cyber threat detected!")
   

About Us

LUNA-ProACT has been released by Phichayut 'Florentin' Sakwiset from WeMake, with contributions from Leon Lukaszewski, Lisa Adolf & Klemens Karboswki.

License

LUNA-ProACT is the property of the Polizei Niedersachsen. It is licensed under the GNU Affero General Public License (AGPL) Version 3 and is available for non-commercial use as an open-source project.

Contact Us

For any further information or inquiries, please reach us at [email protected].

💙 Thanks to the Polizei Niedersachsen

We appreciate the platform provided by Polizei Niedersachsen through the TRUE CRIME HACKATHON 2023, which played a crucial role in the inception and development of LUNA-ProACT.