🤖 RecommendAI: Multi-Recommender System

A premium, production-ready AI recommender system featuring Movie, Song, Book, and Sentiment Analysis modules built with Streamlit, Scikit-Learn, and NLP techniques.

---

🌟 Overview

RecommendAI is a centralized hub for content discovery. Originally an older university capstone project, it has been completely architected and rebuilt into a modern, modular, and performant web application. It uses Content-Based Filtering (Cosine Similarity) and Natural Language Processing (TextBlob) to provide highly accurate recommendations and emotional tone analysis.

Key Highlights

• 10,000+ movie dataset recommendations
• 30,000+ Spotify tracks analyzed
• 6,800+ books indexed
• NLP sentiment classifier trained on IMDB reviews with ~98% accuracy

✨ Core Modules

1. 🎬 Movie Recommender
   • Suggests the top 5 similar movies from a dataset of 10,000+ titles.
   • Integrates live with the TMDB API to fetch high-res posters, ratings, genres, and overviews.
   • Cross-references titles with an IMDB review dataset to display real-time audience sentiment.
2. 🎵 Song Recommender
   • Recommends tracks from a 30,000+ Spotify dataset based on audio features and metadata using TF-IDF.
   • Automatically pulls real album artwork via the iTunes Search API (zero API key required).
3. 📚 Book Recommender
   • Content-based filtering across 6,800+ books.
   • Displays cover art and author ratings perfectly aligned with the cinematic UI.
4. 💬 NLP Sentiment Analysis
   • Background Model: The core sentiment engine was originally modeled using a TF-IDF Vectorizer trained on IMDB movie reviews, achieving a remarkable 98% accuracy in classification.
   • Movie Integration: When a movie is recommended, the system cross-references the title against the IMDB reviews dataset. It aggregates historical reviews for that specific movie and classifies them into Positive, Negative, or Neutral—displaying the exact sentiment counts directly on the movie card to help users gauge audience reaction.
   • Live Analysis Studio: Features a dedicated free-text analysis tool powered by TextBlob. It instantly calculates Polarity (−1.0 to 1.0) and Subjectivity, displaying dynamic emojis and UI color shifts based on the emotional tone of any pasted text.

---

🎨 Premium UI/UX Design

The application distances itself from standard data-science wrappers by implementing a Warm Cinematic Theme inspired by Netflix and IMDB:

• Pure Charcoal Backgrounds (#0d0d0d to #1a1a1a) with no cold blue tints.
• Glassmorphism Elements on result cards.
• Fluid Micro-animations (floating hero icons, pulsing emojis, smooth hover states).
• Custom typography using the Outfit Google Font.

---

🛠️ Architecture & Tech Stack

Category	Technology Used
Frontend Options	Streamlit, HTML/CSS inside Markdown
Data Processing	Pandas, NumPy
Machine Learning	Scikit-Learn (CountVectorizer, cosine_similarity)
NLP	TextBlob
External APIs	TMDB API (Movies), iTunes API (Music)

---

🚧 Challenges Faced & Overcome

Building a robust, deployable recommender system came with several structural challenges:

1. The 763 MB GitHub Constraint

Challenge: The original capstone project relied on a precomputed similarity.pkl matrix that weighed 763 MB, which is strictly blocked by GitHub's 100 MB hard limit. Solution: The architecture was rewritten to compute the cosine similarity matrix on-the-fly. Using CountVectorizer and @st.cache_data, the matrix is generated directly from the movie tags in just ~1.5 seconds on the very first app load, entirely bypassing the need to store or push massive pickle files.

2. Streamlit Security Policies (CSP)

Challenge: Initially, injected HTML <img> tags were used for movie posters, but Streamlit's strict Content Security Policy blocked external image rendering. Solution: The UI was refactored to use native st.image() components styled via CSS targeting the div[data-testid="stImage"] classes, ensuring perfect rendering without security violations.

3. Missing Song Artwork

Challenge: The Spotify dataset lacked album artwork links, and the Spotify API requires complex OAuth tokens. Solution: Implemented a lightweight, robust utility (album_art_fetcher.py) that queries the public iTunes Search API, caching the 600x600 px album covers instantly without requiring an API key.

---

🚀 Running Locally

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/YOUR_REPO_NAME.git
cd YOUR_REPO_NAME

2. Set up the environment

python -m venv .venv

# Windows
.\.venv\Scripts\activate
# Mac/Linux
source .venv/bin/activate

pip install -r requirements.txt

3. Add your TMDB API Key

Create a .streamlit/secrets.toml file in the root directory:

TMDB_API_KEY = "your_v3_api_key_here"

4. Download Datasets

(Since GitHub limits file sizes, the .csv and .pkl datasets are not tracked via Git). Place the following files inside the data/ folder:

• movies_list.pkl
• spotify_songs.csv
• data.csv
• imdb_reviews.csv
• booksummaries.txt

5. Launch the app

streamlit run app.py

---

☁️ Deployment

This project is optimized for deployment on Streamlit Community Cloud:

1. Push your code to GitHub (datasets under 100MB will push successfully).
2. Connect your repository to Streamlit Cloud.
3. Add your TMDB_API_KEY in the Streamlit Cloud Secrets interface.
4. Deploy! The similarity matrix will be computed dynamically on the cloud server.

---

⭐ If you find this project interesting, consider giving it a star!

Built with ❤️ utilizing Python and Machine Learning.