Humor Style Classification XAI Dataset

This repository contains the eXplainable AI (XAI) analysis data for humor style classification, accompanying the paper "Explaining Humour Style Classifications: An XAI Approach to Understanding Computational Humour Analysis" (2024).

Repository Structure

• ├── LIME_plots/ # LIME visualization plots for all analyzed jokes
• ├── linguistic_affective_analysis_json/ # Individual JSON files for each joke
• ├── XAI.xlsx # Combined analysis data
• └── README.md

Dataset Description

The dataset contains XAI analysis for 293 jokes classified into five categories:

• Self-enhancing
• Self-deprecating
• Affiliative
• Aggressive
• Neutral

LIME_plots/

Contains PNG files of LIME (Local Interpretable Model-agnostic Explanations) visualizations for all 293 analyzed jokes. Each plot shows:

• Word-level feature importance
• Contribution to classification decision
• Confidence scores

File naming convention: lime_explanation_joke_{id}.png

linguistic_affective_analysis_json/

Individual JSON files containing detailed linguistic and affective analysis for each joke, including:

• Linguistic features (syllable complexity, semantic conflicts, homonyms, etc.)
• Affective patterns (sentiment, emotion, sarcasm)
• Structural elements (self-references, POS patterns)
• Classification results

File naming convention: detailed_analysis_joke_{id}.json

XAI.xlsx

A consolidated excel file combining all analysis data, with columns:

• joke_id: Unique identifier for each joke
• joke_text: The actual joke content
• true_label: Ground truth humor style
• predicted_label: Model's classification
• confidence_score: Classification confidence
• linguistic_features: Extracted linguistic patterns
• affective_features: Emotional and sentiment analysis
• lime_features: Top contributing features from LIME analysis
• error_analysis: Misclassification details (where applicable)

Requirements

Python 3.7+ pandas matplotlib PIL (Python Imaging Library)

Citation

If you use this dataset in your research, please cite: @inproceedings{kenneth2024explaining, title={Explaining Humour Style Classifications: An XAI Approach to Understanding Computational Humour Analysis}, author={Kenneth, Mary Ogbuka and Khosmood, Foaad and Edalat, Abbas}, booktitle={Journal of Data Mining and Digital Humanities}, year={2024} }

License

This dataset is released under MIT License.You can read more about it at https://opensource.org/license/MIT.

Usage

The data can be loaded and analyzed using standard Python libraries:

import pandas as pd
import json
import matplotlib.pyplot as plt

# Load consolidated data
xai_data = pd.read_excel('XAI.xlsx')

# Load individual JSON analysis
with open('linguistic_affective_analysis_json/detailed_analysis_1.json', 'r') as f:
    joke_analysis = json.load(f)

# View LIME plots
from PIL import Image
img = Image.open('LIME_plots/plots/lime_explanation_0001.png')
plt.imshow(img)
plt.axis('off')
plt.show()