Inside the Screening Room

Discover how three independent AI models analyze the same movie review, explore the technology behind the platform, and learn how The Screening Room brings machine learning and modern software engineering together.

Meet the Critics

The Statistician

Logistic Regression + TF-IDF

The Statistician evaluates a review by measuring how strongly individual words and phrases indicate positive or negative sentiment. It provides a fast, reliable baseline and is highly interpretable.

Strengths
  • Extremely fast inference
  • Easy to interpret
  • Strong traditional machine learning baseline

The Contextualist

BERT

The Contextualist reads every word in relation to every other word, allowing it to understand context, nuanced language, and complex sentence structures.

Strengths
  • Context-aware
  • Handles nuanced language
  • Strong performance on complex reviews

The Sequentialist

Bi-LSTM

The Sequentialist processes each review in order, learning how meaning develops across a sentence. It captures long-term dependencies and contextual flow within the text.

Strengths
  • Understands sentence order
  • Learns contextual sequences
  • Captures long-range dependencies

How It Works

User Review
User submits a movie review.
Model-specific Preprocessing
Each model receives text prepared according to its own preprocessing pipeline.
Parallel Inference
Logistic Regression, Bi-LSTM, and BERT run independently.
Prediction Standardization
Outputs from all models are converted into one consistent response format.
Majority Decision
The application compares predictions and determines the overall verdict.
Frontend Rendering
Predictions, confidence scores, and model insights are displayed simultaneously.

Project Details

Model Accuracy Precision Recall F1 Inference

Metrics are reported using the final evaluation performed on the held-out IMDb test dataset.

IMDb Large Movie Review Dataset

50,000 Reviews
25,000 Positive
25,000 Negative
Balanced Dataset

The models were trained and evaluated using the IMDb Large Movie Review Dataset, a widely used benchmark for binary sentiment classification introduced by Andrew L. Maas and colleagues.

Dataset Attribution

Dataset: IMDb Large Movie Review Dataset

Authors: Andrew L. Maas et al.

Institution: Stanford AI Lab

Official Source: https://ai.stanford.edu/~amaas/data/sentiment/

Frontend

  • HTML
  • CSS
  • JavaScript

Backend

  • Express.js
  • FastAPI

Machine Learning

  • Scikit-learn
  • TensorFlow / Keras
  • Hugging Face Transformers

Deployment

  • Docker
  • Render
  • Neon PostgreSQL
Browser
Express Backend
FastAPI ML Service
Prediction Engine
Three Models
Standardized Response

Behind the Build ⭐

Why FastAPI?

FastAPI powers the machine learning inference service because it provides high-performance APIs while integrating naturally with Python-based ML libraries.

Why Express?

Express separates application logic from machine learning inference by handling validation, routing, persistence, and API orchestration.

Why PostgreSQL?

PostgreSQL stores prediction history and model metadata using a relational schema to demonstrate production-style persistence.

Why Three Models?

The goal is not only to predict sentiment but also to compare how fundamentally different NLP approaches interpret the same review.

Challenge Solution Result
Different preprocessing requirements Model-specific preprocessing pipeline Consistent and reliable predictions
Three incompatible prediction outputs Unified response schema Common API contract across all models
Large BERT model Singleton loading during application startup Reduced inference overhead after initialization
Different confidence formats Standardized prediction response Consistent frontend rendering

Future Improvements

Multilingual Reviews

Planned

Support additional languages beyond English.

Batch Analysis

Planned

Analyze multiple reviews simultaneously.

Additional Transformer Models

Future

Compare more state-of-the-art NLP models.

Mobile Experience

Future

Responsive mobile-first interface.

Advanced Model Interpretation

Research

Expand interpretation techniques for deeper insight into model behavior.