tweet-irony-detection

URL: https://github.com/desh2608/tweet-irony-detection

Description

This project provides a model for irony detection in tweets, specifically developed for the SemEval 2018 Task 3. The model uses a combination of feature generation techniques and classification models to detect irony in tweet text.

Methods

Feature Generation

The model generates features from two different sources:

1. Holographic Embeddings: This method uses circular cross-correlation between tweet text and hashtag vectors. The feature generation script can be found in the holographic.ipynb file.
2. DeepMoji: This feature generation method uses a pre-trained emoji prediction model that has been forked and modified to support Python 3.5+. The script for generating features is located in deepmoji_features.ipynb and must be placed in the DeepMoji/examples directory.

Classification

After feature generation, the XGBoost classifier is employed to classify the generated features. The classification process is implemented in the xgb_classifier.ipynb file.

Results

No specific results or performance metrics are provided in the repository.

Models

The model used in this project is based on XGBoost, a popular gradient boosting algorithm. It is trained using the features generated from the Holographic embeddings and DeepMoji methods.

Dataset

The model relies on tweet data, but the repository does not provide any specific dataset. Users are instructed to use their own data or refer to the SemEval 2018 Task 3 dataset for irony detection.