Overview Of SHapley Additive exPlanations (SHAP) In Natural Language Processing

Introduction

This project is a part of my Thesis "Overview Of SHapley Additive exPlanations (SHAP) In Natural Language Processing", where I primarily worked with Python Notebooks. Throughout the thesis, I dealt with two datasets: the Large Movie Review Dataset, which I used for sentiment analysis, and the IBM Debater: Evidence Sentences Dataset, on which I performed argument mining. Additionally i added my final Thesis as PDF.

German grade 1.0 (very good)

Sentiment Analysis

The Sentiment analysis code can be found in the Sentiment folder. The main notebook for Sentiment Analysis is located inside the pipeline folder and named "Sentiment_Analysis_SHAP". In this notebook, I conducted Sentiment analysis using three different models: Decision Tree, Logistic regression, and BERT. After the training, I locally interpreted certain misclassifications and performed global analysis on the important words.

Argument Mining

Similarly, the code for Argument mining is located inside the Argument folder. The dedicated notebook for Argument Mining can be found inside the pipeline folder, named "Claim_Detection_SHAP". In this notebook, I trained and analyzed the exact same models as in Sentiment Analysis but on a different task, which is Argument mining.

Custom Shap Explainer

For further investigation of certain classifications and to gain a better global overview of the important words, I created a custom Shap explainer. This custom explainer is located inside the custom_shap_explainer folder. It includes custom plots, such as a local word highlighting plot that highlights only the words with the biggest impact, and a global plot that displays a boxplot to provide more information about the distribution of the most important words.

Getting Started

To get started with the code in this repository, follow these steps:

1. Clone this repository to your local machine.
2. Make sure you have the required dependencies installed.
3. Open the respective notebooks for Sentiment Analysis and Argument Mining located in the pipeline folder.
4. Run the code cells in the notebooks to perform the analyses.