Building an NLP Engine to Identify and Fight Psychological Violence

Background

Psychological violence often goes unnoticed, particularly in situations where the severity of harm is minimal or ambiguous. This challenge is compounded in Mexico and Latin America, where cultural and social norms may obscure early warning signs of psychological abuse. Violetta, a technology platform, aims to empower individuals—men and women aged 18 to 45—to recognize and address psychological violence within their social, familial, or professional circles. By providing empathetic, unbiased information, the platform promotes awareness, prevention, and early intervention.

Objective

The primary goal of this project was to develop a robust NLP model capable of identifying patterns of psychological violence in textual data. Additional objectives included:

• Categorizing text inputs into specific types of psychological violence.
• Offering tailored recommendations to users based on their inputs.
• Building a dashboard to analyze data trends and sentiments in real-time.

Approach

Over two months, the Omdena team collaborated to design and implement a solution for psychological violence detection. The approach involved:

• Data Collection: Leveraging both partner-provided datasets and external sources on violence-related content.
• Model Development: Creating a Multi-Label Text Categorization model with near 93% accuracy to classify user inputs into distinct categories of psychological violence.
• Recommendation System: Developing a mechanism to provide actionable recommendations and risk assessments based on the classified data.
• Dashboard Creation: Designing a comprehensive dashboard to monitor sentiments and trending topics, offering valuable insights for prevention strategies.
• Toolset: Advanced NLP techniques and tools such as Python, TensorFlow, and visualization platforms were utilized.

Results and Impact

The project delivered tangible, measurable outcomes:

• A trained NLP model with 93% accuracy to detect psychological violence patterns.
• A categorization system for identifying specific violence types in user-submitted texts.
• Personalized recommendations for users, detailing risk levels based on input analysis.
• A real-time dashboard providing insights into sentiment and trending discussion topics, enabling proactive measures.
• These advancements enhance Violetta’s ability to address psychological violence effectively, empowering individuals to recognize and respond to harmful behaviors.

Violeta about the AI Challenge results

Future Implications

The findings from this project pave the way for significant advancements in the fight against psychological violence. Future steps include:

• Integrating the NLP model into conversational AI systems like chatbots, enabling real-time support for users.
• Expanding the model’s capabilities to detect additional forms of abuse or violence.
• Informing public policies and awareness campaigns by leveraging insights gained from dashboard data.
• Enhancing international applications of the model to broaden its impact globally.