Data Annotation System for Valence-Arousal-Irony

Introduction

A PHP + MySQL website for annotating PTT comments on three sentiment dimensions (valence, arousal, irony) plus an incomplete-sentence flag. Annotators log in, read a post, and score each comment on a 0–5 scale; progress is saved per user.

The irony annotations collected with this system and its 2021 extension became the CDVAI dataset:

• A Dimensional Valence-Arousal-Irony Dataset for Chinese Sentence and Context (ROCLING 2022)
• A Chinese Dimensional Valence-Arousal-Irony Detection on Sentence-level and Context-level Using Deep Learning Model (IJCLCLP Vol. 27, No. 2, 2022)
• Paper walkthroughs on my blog: ROCLING dataset paper · IJCLCLP detection paper

The Heroku demo was shut down when Heroku ended its free tier.

Requirements

• Python >= 3 with pandas and MySQL Connector/Python
• XAMPP (Apache + MySQL)
• PHP 7.2

Getting Started

1. Start Apache and MySQL in XAMPP.
2. Copy this folder under htdocs/ (or /var/www/html/ on a Linux server).
3. Open phpMyAdmin (http://localhost:8080/phpmyadmin/, adjust the port) and create an empty database named VAI.
4. Set the database credentials in config.php.
5. Create the annotator accounts: copy sql_processing/data/user.example.csv to sql_processing/data/user.csv and fill in one name,number,password row per annotator. user.csv is git-ignored, so real accounts never get committed.
6. Load the schema, accounts, and posts:

   cd sql_processing
   python ./sql_processing.py
   

7. Open http://localhost:8080/annotation_system/ and log in with one of the accounts.

Deploying on Heroku

The original deployment used the heroku-php-apache2 buildpack: a composer.json pinning "php": "7.2.28", a Procfile containing web: heroku-php-apache2, and the first three lines of index.php uncommented to load Composer's autoloader. Any PHP host with MySQL works the same way.