Irony-Detection-Master-Thesis-Files

URL: https://github.com/ThorTheStone/Irony-Detection-Master-Thesis-Files

Description: Files related to the “Automated Moderation: Detecting Irony in a Norwegian Facebook Comment Section using a Longformer Transformer Model with a Context Encoded Dataset” master thesis.

Methods

• Model:
  • Longformer Transformer Model: A transformer-based model for handling long sequences, tailored to detect irony in Facebook comment sections.
  • Context Encoded Dataset: Dataset includes context encoding to enhance the model’s understanding of irony within conversation threads.
• Approach:
  • Automated moderation using advanced machine learning techniques for detecting irony in comments.
  • Data Processing: Textual data from Facebook comments is pre-processed before being fed into the model.

Results

• The performance of the model is evaluated using various metrics for classification.
• Results focus on accuracy and ability to detect subtle irony in conversational contexts, particularly within Norwegian-language datasets.

Dataset

• Dataset Used:
  • The dataset consists of Norwegian Facebook comments, with an emphasis on ironic comments.
  • Context Encoding: The dataset incorporates conversation threads to provide additional context to the model.
• Data Processing: Pre-processing is required to clean and format the data before training with the Longformer model.