Inspiration
In today’s digital world, social media, advertisements, and online content often use subtle emotional manipulation and persuasive tactics. I wanted to build a tool that reveals hidden manipulative language, helping users understand how text can influence emotions and decisions. This was inspired by the need for digital literacy and critical thinking in the age of information overload.
What it does
Manipulation X-Ray analyzes any text and detects: Manipulation categories: Fear, Urgency, Guilt, Exaggeration, Clickbait Persuasion tactics: Authority, Scarcity, Reciprocity, Bandwagon Emotional cues: Fear, Guilt, Joy, Anger, Sadness
It highlights manipulative words, provides overall manipulation scores, and displays interactive visualizations like heatmaps, radar charts, and bar charts. Users can also see a neutral rewrite suggestion for manipulative sentences.
How we built it
Text Processing: Input text is split into sentences using NLTK’s tokenizer. Manipulation Detection: Keyword-based classifiers detect fear, urgency, guilt, exaggeration, and clickbait. Persuasion & Emotion Analysis: Scores are computed for persuasion tactics (authority, scarcity, reciprocity, bandwagon) and emotional cues (fear, guilt, joy, anger, sadness). Visualization: Interactive sentence-level heatmaps, radar charts, and bar charts display manipulative patterns. Interactive Demo: ipywidgets allow users to input text and instantly see analysis results in Colab.
Challenges we ran into
Widget metadata issues: Colab widgets caused “Invalid Notebook” errors on GitHub, solved by cleaning metadata. Balancing simplicity and wow-factor: Needed to combine scores, visualizations, and sentence highlights in a single output block. Designing interactive charts: Ensuring charts are clear, readable, and visually impressive for judges.
Accomplishments that we're proud of
Combining text analysis, emotion detection, and interactive visualizations in a single demo. Designing a clear, wow-factor output block that shows scores, charts, and sentence-level insights at once. Making the project fully runnable in Colab, so judges can interact with it without setup.
What we learned
Through this project, I gained hands-on experience in Natural Language Processing (NLP), text tokenization, sentiment and emotion detection, and interactive data visualization. I also improved my skills in Python, NLTK, Plotly, Matplotlib, Seaborn, and ipywidgets, learning how to build real-time, interactive demos for end users.
What's next for Manipulation Detector
Integrate Hugging Face zero-shot classifiers for more accurate manipulation detection. Add automatic neutral rewrites for manipulative sentences. Deploy a web version using Streamlit or Gradio for a polished, interactive demo outside Colab.
Log in or sign up for Devpost to join the conversation.