Text classification of customer inquiries via e-mail

Armin WEIGOLD

Master's thesis

Abstract:

This work addresses a problem of text classification. The research question focuses on identifying the optimal workflow to solve the given problem. The dataset consists of roughly 430,000 labeled e-mails. The problem is tackled in two steps, namely vectorizing the text and applying a classification algorithm. Several algorithms, including Word2vec and Tf-idf for vectorization, and Random Forest, Support Vector Machine, Graph Neural Network, and Feed-Forward Neural Network for classification, were evaluated. The results show that Word2Vec performed well, while Tf-idf had too high memory demands. In terms of classification, the Feedforward Neural Network achieved the highest F1 scores of 0.89- 0.90 depending on the trial, followed by Random Forest and Support Vector Machine with F1 scores of 0.87-0.89, while the graph neural network achieved F1 scores of 0.80-0.87.

Language used: English

Date on which the thesis was submitted / produced: 6. 9. 2023

Thesis defence

• Supervisor: prof. Dr. Andreas Fischer

UNIVERSITY OF SOUTH BOHEMIA IN ČESKÉ BUDĚJOVICE

Faculty of Science

Master programme / field:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science