Skip to content

README.md

Latest commit

 

History

History
209 lines (152 loc) · 7.83 KB

README.md

File metadata and controls

209 lines (152 loc) · 7.83 KB

🧘 MindY - Human-Centered Exploration of Mindfulness Content Discovery 🎥✨

MindY is a concept and exploration project that examines how people search for and engage with mindfulness and relaxation content. It uses natural language processing and information retrieval techniques as a lens to explore personalization, language, and user intent, rather than to deliver a production-ready wellness tool.

This project integrates YouTube metadata retrieval and natural language processing (NLP) with TF-IDF vectorization and cosine similarity to rank and recommend relevant relaxation and mindfulness content.

Context & Intent

MindY was developed as part of a learning journey focused on human-centered questions around content discovery. Its purpose is to explore how people search for and engage with mindfulness and relaxation content, rather than to deliver a production-ready tool.

The project focuses on experimentation, reflection, and design considerations, with attention to language, intent, and ethical use of technology in wellbeing contexts.

📑 Table of Contents

  1. Project Overview
  2. Key Features
  3. How It Works
  4. Project Structure
  5. Requirements
  6. Usage
  7. Why This System?
  8. Next Steps

📖 Project Overview

MindY consists of two core functionalities:

  1. 🧠 AI-Powered Chatbot for Personalized Techniques

    • Uses GPT-4 to generate actionable wellness techniques based on user queries.
    • Allows users to select wellness categories for more refined recommendations.

  2. 🎥 Intelligent YouTube Video Search

    • Retrieves relevant wellness videos from YouTube API.
    • Ranks videos by semantic similarity using TF-IDF & cosine similarity.
    • Ensures users get high-quality, meaningful video recommendations.

✨ Key Features

🎯 Personalized Recommendations

  • Uses GPT-4 to suggest specific techniques tailored to user queries.
  • Example: A query like "How can I reduce stress?" may generate:
    • Guided Breathing Exercise
    • Mindfulness Meditation for Anxiety Relief

🔍 Smart Video Search

  • Queries YouTube API and ranks videos by similarity to the user's query.
  • Prevents generic recommendations by matching video descriptions to techniques.

📂 Category-Based Suggestions

  • Users can select predefined wellness categories, such as:
    • 🧘 Mindfulness & Meditation
    • 💨 Breathing Exercises
    • 🤸 Somatic Practices

🤔 How It Works

Query Processing Pipeline

  1. 💬 User Input

    • The app collects a query (e.g., "How do I sleep better?").
    • If a category is selected, the response is focused on that domain.

  2. 🧠 AI Response Generation

    • GPT-4 generates a personalized list of techniques.
    • Extracts keywords and themes from the response.

  3. 🔗 YouTube Search & Ranking

    • The system queries YouTube API for videos using refined keywords.
    • Uses TF-IDF & cosine similarity to re-rank videos based on relevance.

  4. 📜 Results Display

    • The app presents AI-generated techniques alongside top-ranked videos.

📂 Project Structure

Key Files & Directories

  • app.py → Main Streamlit app (chatbot & video recommendations)
  • requirements/ → Contains dependencies for different functionalities:
    • mindy_app.txt → Dependencies for Streamlit app
    • genai.txt → Dependencies for query simulation & clustering
    • nlp.txt → Dependencies for data analysis
  • simulate_queries/ → Query simulations & GPT data extraction
    • Contains the following key files:
      • app_logic.py: Implements core logic for GPT-4 recommendations, YouTube video fetching, ranking, and query processing.
      • simulate_queries.py: Automates the simulation of user queries using app_logic.py to extract GPT-recommended techniques and descriptions. Outputs results to simulation_results.csv.
  • data_analysis/ → NLP analysis & clustering scripts:
    • Contains scripts for text preprocessing, clustering, and visualization of results.
    • Includes simulated_results.csv, which holds the output of simulated user queries for analysis.
  • MindyPresentation.pdf → Project presentation
  • README.md → Project documentation

⚙️ Requirements

  1. API Keys:

    • YouTube API Key (store your Key in a .env file as YOUTUBE_API_KEY).
    • OpenAI API Key (store your Key in the .env file as OPENAI_API_KEY).

  2. Dependencies:

    • Python libraries: streamlit, openai, scikit-learn, google-api-python-client, python-dotenv.

  3. Install Dependencies:

    • To keep dependencies isolated and avoid conflicts, create separate virtual environments for each functional area.

    🟢 MindY App (Streamlit)

    1. Create a virtual environment for the Streamlit app:
    python3 -m venv mindy_app_env
source mindy_app_env/bin/activate  # On Windows, use mindy_app_env\Scripts\activate
    1. Install the dependencies:
    pip install -r requirements/mindy_requirements.txt

    🟡 Simulating Queries (GPT Data Extraction & Clustering)

    1. Create a virtual environment for query simulation and clustering:
    python3 -m venv genai_env
source genai_env/bin/activate  # On Windows, use genai_env\Scripts\activate
    1. Install the dependencies:
    pip install -r requirements/genai.txt

    🔵 Data Analysis (Clustering & NLP Processing)

    1. Create a virtual environment for data analysis:
    python3 -m venv nlp_env
source nlp_env/bin/activate  # On Windows, use nlp_env\Scripts\activate
    1. Install the dependencies:
    pip install -r requirements/nlp.txt

🚀 Usage

1️⃣ Run the Application

  • Activate the virtual environment for the MindY App:
source mindy_app.env/bin/activate  # On Windows: mindy_app.env\Scripts\activate
  • Launch the chatbot and video recommender:
streamlit run app.py

2️⃣ Input Queries

  • Enter a wellness-related query (e.g., "How can I improve my focus?").
  • Optionally, select a category for more refined suggestions.

3️⃣ View AI-Powered Recommendations

  • The app displays customized techniques & top-ranked YouTube videos.

❓ Why This System?

✅ Traditional YouTube searches lack personalization.

This system improves recommendations by:

  • 🎯 Understanding query intent (semantic search, not just keywords).
  • 🛠️ Filtering & ranking YouTube videos for better quality.
  • 💡 Providing AI-powered guidance, not just links.

🛠️ Next Steps

🚀 Integrate Clusters into GPT’s Workflow

  • Leverage clustered techniques to provide more cohesive recommendations.
  • Map user queries to cluster themes for more personalized outputs.

🔧 Enhance Query Preprocessing

  • Improve text preprocessing by refining techniques like removing stop words and applying lemmatization.

🎯 Refine Prompt Design

  • Shift from a narrow focus on Mindfulness and Relaxation to a broader range of well-being areas.
  • Expand prompts to include diverse domains like emotional management and cognitive behavioral therapy (CBT).

📚 Broaden Categories

  • Introduce new categories (e.g., journaling prompts) to generate more varied and meaningful recommendations.

💡 Experiment with Free Models

  • Simulate queries using a generalized system, testing a free model like DeepSeek.

🔄 Evaluate Clustering Impact

  • Repeat clustering processes and analyze how the new implementations influence results.