Sarcasm-Detection

Table of Contents

• Project Overview
  • Project Background
  • Key Features
  • Technology Stack
• Environment Requirements
  • Python Version
  • Dependencies
• Quick Start
  • Configuration File
    • How to Edit the Configuration File
  • Data Preparation
    • Dataset Source
    • Contextual Data
  • Training the Model

Project Overview

Project Background

Sarcasm is a prevalent rhetorical device encountered in social media, news comments, or everyday conversations. However, it presents a considerable challenge for Natural Language Processing (NLP) tasks. This project aims to develop an efficient and accurate sarcasm detection model to understand textual content more accurately in various applications such as sentiment analysis and public opinion monitoring.

Key Features

• Sarcasm Detection: Utilizes advanced deep learning algorithms to classify the sarcastic sentiment of the input text.
• Contextual Understanding: Through interaction with relevant external knowledge bases (like Wikipedia), the model can grasp the text's context more accurately.
• Multilevel Analysis: Beyond basic text analysis, the model employs advanced features like multi-head attention mechanisms and bi-directional LSTM to capture complex relationships within the text.

Technology Stack

• PyTorch: Used for model building and training.
• Flair: Employed for text classification and natural language processing tasks.
• Transformer: Word embeddings based on the Roberta model.
• AdamW: Serves as the model's optimizer.
• OneCycleLR: Utilized for learning rate scheduling.

Environment Requirements

Python Version

This project is implemented using Python 3.7

Dependencies

• PyTorch 1.13.1

Additional package dependencies can be found in the requirements.txt file and installed using pip:

pip install -r requirements.txt

Quick Start

Configuration File

The configuration for the model, optimizer, and other settings can be found in the YAML file (iron.yaml). Make sure to go through the configuration to understand the different parameters.

How to Edit the Configuration File

To edit the configuration file, simply open iron.yaml in a text editor and modify the values according to your needs. For example, to change the learning rate:

train:
  learning_rate: 2e-5  # Change this value

Data Preparation

Dataset Source

The dataset used for this project originates from the SemEval 2018 competition. The original dataset can be downloaded from here

Contextual Data

In this project, contextual information is added from Wikipedia, New York Times, and BBC. The processed dataset with context is available here

Training the Model

Run the training script:

python train_iron.py