ItemConference Paper

Fairness-Aware and Interpretable Depression Detection on Social Media Using BERT with Gender Bias Mitigations

Title
Fairness-Aware and Interpretable Depression Detection on Social Media Using BERT with Gender Bias Mitigations
Creator
George, Agnus Maria; Kavitha, R.; Kumar, Dalvin Vinoth
Description
Reddit and similar social media platforms offer substantial information regarding mental health issues. The automatic detection of depression raises different issues pertaining to fairness and transparency. This paper presents a Fairness Aware and Interpretable Depression Detection framework that utilizes BERT and incorporates an explicit gender bias mitigation mechanism. Data were obtained from gender-specific forums on Reddit. The Mistral language model based classifier was used to set a high confidence threshold, which helped in inferring gender labels while both depressed and non-depressed were among the patients assigned the labels. A balanced dataset with four groups (Depressed-Male, Depressed-Female, NonDepressed-Male, NonDepressed-Female) was prepared. Two pipelines were carried out where one involved a baseline BERT classifier while the other employed a fairness aware BERT model that incorporated gender embeddings during the training phase. The models were assessed using accuracy, precision, recall, F1 score, and confusion matrices and the fairness metrics applied were Demographic Parity Difference (DPD) and Equal Opportunity Difference (EOD). To enhance the model's reasoning transparency, SHAP was applied due to its capability to provide clear and comprehensive explanations. The results indicated that the fairness centered model effectively reduced gender biasness and equalized error rates among the different groups without losing its original accuracy. The essential point is that the model had learned to give precedence to clinical indicators over gender specific language. This study suggests a roadmap for the creation of ethical AI by combining fairness, interpretability and high performance into a seamless framework. 2025 IEEE.
Source
2025 17th IEEE International Conference on Computational Intelligence and Communication Networks, CICN 2025;pp.1973-1979
Date
01-01-2025
Publisher
Institute of Electrical and Electronics Engineers Inc.
Subject
BERT; Depression detection; Fairness aware machine learning; Gender bias mitigation; Interpretability; Social media
Coverage
George A.M., CHRIST (Deemed to Be University), Bangalore, India; Kavitha R., CHRIST (Deemed to Be University), Bangalore, India; Kumar D.V., CHRIST (Deemed to Be University), Bangalore, India
Rights
Restricted Access; Hardcopy may be available in the library
Relation
ISBN: 979-833158733-8;
Format
online
Language
English
Type
Conference paper

Citation

George, Agnus Maria; Kavitha, R.; Kumar, Dalvin Vinoth, “Fairness-Aware and Interpretable Depression Detection on Social Media Using BERT with Gender Bias Mitigations,” CHRIST (Deemed To Be University) Institutional Repository, accessed June 18, 2026, https://archives.christuniversity.in/items/show/25788.

Output Formats