NLP & AI Projects:

NLP-20260313

Quantified Explainability and Robustness Analysis of Transformer-Based Bug Detection Models

by Debargha Ghosh, Eve Thullen, and Emmanuel Udoh Ph.D.; THULLEN RESEARCHLAB, 02/02/2026

Unplanned equipment failures in manufacturing systems lead to production downtime, increased operational costs, and safety risks. While predictive maintenance techniques have advanced significantly, much of the existing work focuses on binary failure detection and provides limited insight into specific failure mechanisms. This paper presents a multi-class predictive modeling approach for manufacturing equipment maintenance systems that aims to identify distinct failure types using operational sensor data.

The study formulates failure type prediction as an imbalanced multi-class classification problem representative of real-world industrial environments, where failure events are rare compared to normal operation. Model performance is evaluated using imbalance-aware metrics to ensure reliable assessment across both dominant and minority failure classes. The results demonstrate that the proposed approach can effectively distinguish major mechanical and thermal failure types despite severe class imbalance. These findings highlight the importance of multi-class failure prediction for enabling more targeted maintenance decisions and improving the reliability of manufacturing equipment.

a close up of a computer screen with code on it

See Paper Details

Cite: [1] Debargha Ghosh, Eve Thullen Ph.D, Emmanuel Udoh Ph.D, "Quantified Explainability and Robustness Analysis of Transformer-Based Bug Detection Models," International Multidisciplinary Research Journal Reviews (IMRJR), 2026, DOI 10.17148/IMRJR.2026.030304

NLP-20250701

Social Media Sentiment Analysis Model for Stock Market Trends

by Samuel Adebayo, and Guru Batchu, University of the Cumberlands, 07/01/2025

This study explores the predictive power of social media sentiment analysis on short-term financial market movements. Social media platforms, particularly Reddit, significantly influence market dynamics through the rapid dissemination of retail investor sentiment. Leveraging sentiment analysis, technical indicators, and advanced machine learning techniques such as Random Forest and LSTM networks, we assessed Reddit-derived sentiment's predictive capabilities.

Our comprehensive analysis revealed weak correlations between social sentiment scores and stock returns, primarily due to high volatility and sentiment noise. Enhanced modeling approaches incorporating technical indicators showed minor improvements, emphasizing the market predictions' multifaceted nature. Practical implications suggest caution in solely relying on sentiment analysis, recommending a combined approach with traditional market indicators. The study underscores the necessity of integrating broader macroeconomic indicators and detailed sentiment metadata for more accurate predictive results.

Request Paper Details

NLP-20250702

Social Media Text Analysis Model for Mental Health Prediction

by Shivakshi Chauhan, and Hissey Lama, University of the Cumberlands, 07/01/2025

The problem of depression, anxiety and stress is growing alarmingly everyday in the world. Social media is another platform that people regularly use to express their emotions, and it is possible to research them to reveal any probable mental issues.

The project proposes a machine learning solution in natural language processing (NLP) to extract mental health signals in social media articles, particularly that of the Reddit and Twitter data sets.

The project utilizes the classical and advanced models such as TF-IDF with logistic regression, Random Forests, and a BERT-based classifier with the addition of sentiment features. High degree of accuracy has been shown in the use of experimental findings, where the BERT model achieved the level of 99 percent as far as accuracy is concerned, a fact that has in turn affirmed the prospect of digital tools in the facilitation of early intervention and detection in mental health.