MetaOpt-CNN for Twitter Sentiment Classification

Metaheuristic hyperparameter optimization for CNN-based Twitter sentiment analysis. This project provides a modular Python package to preprocess tweet data, build and evaluate CNN models, and optimize both model and optimizer hyperparameters using multiple metaheuristic algorithms.

Features

• Two-phase optimization:
  • Phase 1: Optimize CNN hyperparameters (HC, PSO, SA, Tabu)
  • Phase 2: Optimize metaheuristic hyperparameters (ACO, Bat Algorithm, internal tuning)
• Tokenization and preprocessing pipeline (NLTK)
• Modular optimizers (each algorithm in its own module)
• CLI with configurable hyperparameter ranges and algorithm settings
• Unit tests (pytest) and CI workflow (GitHub Actions)
• Docker support for reproducible runs
• JSON logging for experiment tracking and analysis tools

Quick Start

1. Clone the repo: git clone https://github.com/aadhamashraf/MetaOpt-CNN-for-Twitter-Sentiment-Classification.git
2. Install dependencies: pip install -r requirements.txt
3. Run Phase 1 (example): python main.py --phase 1 --csv path/to/twitter_training.csv
4. Run Phase 2 (example): python main.py --phase 2 --csv path/to/twitter_training.csv

CLI

• --phase [1|2] : Choose optimization phase
• --csv PATH : Path to dataset CSV (defaults to twitter_training.csv)
• --search-space-file PATH : JSON file with hyperparameter ranges
• --pso-particles N
• --pso-iters N
• --hc-iters N
• --sa-iters N
• --ts-iters N
• --aco-ants N
• --aco-iters N
• --ba-bats N
• --ba-iters N

(See main.py --help for full list)

Project Structure

• src/
  • config.py
  • logger.py
  • data.py
  • cnn.py
  • optimizers/
    • init.py
    • aco.py
    • bat.py
    • pso.py
    • hill_climb.py
    • simulated_annealing.py
    • tabu.py
    • tuner.py
  • run_phase1.py
  • run_phase2.py
  • analyze.py
• tests/
• Dockerfile
• requirements.txt
• README.md

Logs are written to optimization_log.json by default.