Explainable and Fair Resume Screening with BERT Embeddings and Tabular Attention Networks

Algaskhanpet, Preethi (2025) Explainable and Fair Resume Screening with BERT Embeddings and Tabular Attention Networks. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (907kB) | Preview

Abstract

AI powered hiring system are widely used to automate the resume screening, but there is major issue faced by them is transparency and hidden bias within the historical recruitment data. The research proposes a hybrid model emphasizing explainability that combines BERT’s strengths in extracting semantic representations from textual skills data with the TabNet architecture for capturing interactions between tabular features with the help of sequential attention. Public available synthetic hiring dataset was prepared based on the requirements, and then feature engineering techniques were done, including skill counts, experience level of the employee, cultural fit scoring, and numerical standardisation. Multiple models were included in the evaluation, such as Logistic Regression, Support Vector Machine (SVM), Decision Tree, Deep Neural Network and a hybrid BERT + TabNet model, with SHAP for interpretability. By the hybrid approach, BERT generated embeddings were combined into the TabNet model to enhance contextual understanding of the candidate profiles. This hybrid model achieved an accuracy of 87.2%, F1-score of 0.565, and ROC-AUC of 0.911 thus it performed better than the traditional ML models and DL model and at the same time, offered a great level of transparency through SHAP explanations. Even-though the Decision tree and Logistic regression had high accuracy, it lacked interpretablitiy , So that it is not suitable for real -world hiring contexts where transparency and fairness is important. The hybrid model provides a solution by combining the powers of semantic understanding, interpretable feature selection, and explainable output visualization, thus granting it the robustness and transparency of an ethical AI-driven recruitment framework.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Mattos, Agatha UNSPECIFIED
Uncontrolled Keywords:	AI hiring systems; Explainable AI; BERT embeddings; TabNet; SHAP values; Resume screening; Fairness in machine learning; Human-centered AI; Hybrid deep learning model
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence H Social Sciences > HD Industries. Land use. Labor > HD28 Management. Industrial Management > Human Resource Management Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HD Industries. Land use. Labor > HD28 Management. Industrial Management > Human Resource Management > Recruitment
Divisions:	School of Computing > Master of Science in Artificial Intelligence for Business
Depositing User:	Ciara O'Brien
Date Deposited:	02 Sep 2026 11:31
Last Modified:	02 Sep 2026 11:31
URI:	https://norma.ncirl.ie/id/eprint/9773

Actions (login required)

View Item