Duolingo's AI Stance: A Sentiment Analysis of Reddit's Response

Strategic shifts are common in business, but measuring audience reactions at scale demands advanced NLP techniques. This project analyzes 11,000+ Reddit posts/comments to uncover emotional patterns around Duolingo's move toward AI.

Project Overview

On April 28, 2025, Duolingo surprised users by announcing its decision to become an AI-first company. By combining Reddit data with state-of-the-art NLP techniques, this project reveals:

• How public emotions evolved around the AI announcement.
• The dominant sentiments expressed during critical phases.
• Whether Duolingo's clarification successfully shifted the narrative.

As a data science enthusiast and longtime Duolingo user (with a 1,775-day streak to prove it!), this project gave me a chance to apply NLP to real-world reactions during a corporate pivot. It highlights the crucial role of data science in extracting valuable insights from unstructured text data.

Key Findings

• Emotional Surges: The "AI-first" announcement triggered sharp spikes in curiosity, confusion, and disappointment on Reddit, far surpassing earlier peaks during the "Duo is dead" campaign.
• Sustained Concern: While Duolingo's CEO clarified the stance, negative sentiments persisted, indicating continued community concerns.
• Core Concerns: Discussions centered around the impact of AI on human jobs and contractors, financial motivations, and direct effects on user engagement.

Repository Structure

sentiment-analysis-duolingo/
├── analysis/
│   └── analysis.ipynb                            # Jupyter Notebook with full workflow
├── data/
│   ├── raw/
│   │   ├── duo_ai_raw.csv                        # Raw Reddit posts and comments
│   │   └── duo_mask.png                          # Duo-shaped mask for word cloud
│   └── processed/
│       ├── duo_ai_cleaned.csv                    # Text after preprocessing
│       ├── duo_ai_with_sentiment_raw.csv         # Sentiment scores (raw)
│       └── duo_ai_with_sentiment.csv             # Filtered sentiment labels
├── figures/
│   ├── Fig1_emotion_confidence_distribution.png  # Distribution of Emotion Confidence Scores
│   ├── Fig2_weekly_emotion_trends.png            # Weekly Counts of Dominant Emotions
│   ├── Fig3_opposing_emotions_trend.png          # Approval vs Disapproval & Admiration vs Disappointment
│   └── Fig4_wordcloud_spacy.png                  # Post-Announcement Word Cloud
├── .gitignore                                    # Git ignore rules
├── environment.yml                               # Environment specifications
├── LICENSE.txt                                   # Project license
└── README.md                                     # This README file

Tech Stack

• Data Collection: PRAW (Python Reddit API Wrapper)
• Data Handling: Pandas, NumPy, datetime
• NLP & Text Processing: Hugging Face Transformers (joeddav/distilbert-base-uncased-go-emotions-student), NLTK, SpaCy, emoji
• Visualization: Matplotlib, Seaborn, WordCloud, PIL

How to Reproduce

1. Clone the Repository

   git clone https://github.com/MathRC/sentiment-analysis-duolingo.git
   cd sentiment-analysis-duolingo

2. Set Up the Environment

   conda env create -f environment.yml
   conda activate sentiment-analysis-Duolingo

   Install SpaCy and NLTK data:

   python -m spacy download en_core_web_sm
   python -c "import nltk; nltk.download('stopwords'); nltk.download('wordnet'); nltk.download('punkt')"

3. Configure Reddit API Credentials

   Create a config.ini with your Reddit API credentials in the project root:

   [reddit]
   client_id     = YOUR_CLIENT_ID
   client_secret = YOUR_CLIENT_SECRET
   username      = YOUR_REDDIT_USERNAME
   password      = YOUR_REDDIT_PASSWORD

4. Run the Analysis

   cd analysis
   jupyter notebook analysis.ipynb

   Follow the notebook to:

   • Collect data via PRAW
   • Preprocess text data
   • Perform sentiment analysis
   • Generate visualizations saved in figures/

Visual Highlights

Here are some of the key visualizations from the project that illustrate the findings:

Figure 2: Weekly counts of the top 5 dominant emotions expressed on Reddit over time, with the "AI-first" announcement date marked.

Figure 3: Comparison of Approval vs. Disapproval and Admiration vs. Disappointment shows negative sentiment persisted even after the May 23 CEO clarification.

Figure 4: Frequently discussed topics in the post-announcement period. Dominant themes like "human," "AI," and "job" reflect concerns about automation and workforce impact.

License

This project is licensed under the MIT License. See LICENSE.txt for details.

Acknowledgments

A special thank you to Duolingo for continuing to make language learning fun and for providing such an interesting case study for sentiment analysis! And, of course, to the incredible open-source community for the powerful NLP and data science tools.

Connect: LinkedIn • GitHub
📧 Email: matt21rc@gmail.com

Open to collaboration and feedback!