Reel Feel: An Emotion-Aware Movie Recommender Using Multi-Source Sentiment Analysis

We've all been there—scrolling endlessly through Netflix, not sure what to watch. Traditional recommendation systems suggest movies based on what you've watched before or what's popular, but they miss a crucial element: how you're feeling right now?

Are you in the mood for something uplifting? A tear-jerker? An adrenaline-pumping thriller? Your emotional state matters, but most recommenders don't account for it.

That's why we built Reel Feel—an intelligent movie recommendation system that matches films to your current emotional preferences using multi-source sentiment analysis.

The Approach: Emotions as Data

The core idea is simple: What if we could quantify the emotional content of movies and match them to how you're feeling?

Instead of relying on genre tags or collaborative filtering alone, we analyzed emotions from three different perspectives:

By combining these three sources, the system creates a comprehensive emotional profile for each movie.

• Professional Critics - How do film reviewers describe the emotional tone?

• Audience Reviews - What emotions do viewers experience?

• Movie Trailers - What emotions does the visual/audio content evoke?

NLP Pipeline: From Text to Emotions

The emotion detection pipeline uses state-of-the-art transformer models from HuggingFace:

• Model: distilbert-base-uncased-finetuned-sst-2-english

• Purpose: Classify overall sentiment (positive/negative)

• Output: Sentiment score from -1 (negative) to +1 (positive)

Sentiment Analysis

Emotion Classification

• Model: j-hartmann/emotion-english-distilroberta-base

• Purpose: Multi-label emotion detection

• Output: Scores for 7 emotions (anger, disgust, fear, joy, neutral, sadness, surprise)

The Recommendation Algorithm: How it feels for you?

This project represents everything I love about data science: taking a real-world problem (finding the right movie for your mood), applying rigorous technical methods (NLP, ML, transfer learning, sentiment analysis), and delivering it through an intuitive user experience.

More importantly, it demonstrates that machine learning doesn't have to be a black box. By surfacing emotion scores, sentiment gaps, and match percentages, users understand why they're seeing each recommendation. This transparency builds trust and makes the system feel collaborative rather than algorithmic.

If you're a fellow data scientist, ML engineer, or just someone who loves movies, I'd love to hear your feedback. What emotions would you add? How would you weight the sources differently? What movies are you excited to discover?