Crisis Detection using NLP, Sentiment Analysis & Geospatial Mapping

CrisisWatch-NLP is a social media analysis tool. The goal is to build a pipeline that detects, analyzes, and visualizes crisis-related discussions (e.g., suicidality, mental health distress, substance use) from Reddit using advanced NLP techniques and geospatial tools.

• 🔍 Extract crisis-related posts from Reddit based on keyword filters and location cues
• 💬 Analyze sentiment using VADER
• 🧠 Assess crisis risk levels using BERT-based semantic similarity
• 🗺️ Visualize regional distress patterns on an interactive map

• Collected posts via Reddit’s praw API from mental health subreddits
• Filtered posts using a keyword list (e.g., “i want to die”, “relapse”)
• Cleaned text using nltk (stopwords, punctuation, emojis removed)
• Output: raw_reddit_posts.csv, cleaned_reddit_posts.csv

• Applied VADER sentiment scoring (Positive, Neutral, Negative)
• Embedded post text and high-risk phrases using BERT (sentence-transformers)
• Calculated semantic similarity to label risk levels:
  • 🟥 High-Risk
  • 🟧 Moderate Concern
  • 🟩 Low Concern
• Output: reddit_bert_risk_labeled.csv, distribution plots

• Extracted city/state names using spaCy NER (GPE labels)
• Geocoded locations with geopy.Nominatim
• Created interactive Folium heatmap of distress discussions
• Displayed Top 5 crisis-prone locations
• Output: crisis_heatmap.html

1. Open the notebook in Google Colab
2. Run code cells step-by-step:
   • Extract posts
   • Clean text
   • Run sentiment & BERT classification
   • Extract and map locations
3. Download outputs:
   • CSV files
   • crisis_heatmap.html

/crisiswatch-nlp/
├── raw_reddit_posts.csv
├── cleaned_reddit_posts.csv
├── reddit_bert_risk_labeled.csv
├── crisis_heatmap.html
├── README.md
└── [Colab Notebook].ipynb