Transformer-Based Multimodal Intelligence for Software Defect Detection: A Cloud-Native LLM Framework with Explainable AI for Digital Infrastructure Maintenance
This paper presents a cloud-native, transformer-based multimodal intelligence framework that integrates Large Language Model semantic encoding with deep neural learning to enable automated defect prediction and proactive maintenance of large-scale digital infrastructure.
Abstract
Modern digital infrastructure generates heterogeneous, multimodal software artifacts encompassing structured code metrics, unstructured textual data such as commit logs and inline comments, and dynamic runtime signals whose complexity renders traditional defect detection approaches increasingly inadequate. This paper presents a cloud-native, transformer-based multimodal intelligence framework that integrates Large Language Model (LLM) semantic encoding with deep neural learning to enable automated defect prediction and proactive maintenance of large-scale digital infrastructure. The proposed system employs a sentence-transformer encoder (all-MiniLM-L6-v2) to process multimodal software artifact data—including serialized structured metrics and available textual fields—into dense 384-dimensional semantic embeddings. These embeddings are subsequently refined through a hierarchical multi-layer perceptron (MLP) deployed on a scalable cloud architecture for real-time inference. Evaluated on a real-world dataset of approximately 60,000 software modules, the framework achieves 99.72% accuracy, 100% precision, and an ROC-AUC of 0.9998, substantially outperforming baseline models including Random Forest, XGBoost, LSTM, and standalone MLP architectures. To address potential concerns regarding result validity, we conducted repeated experiments with five different random seeds (42, 0, 1, 7, 123) and ten-fold stratified cross-validation, confirming that performance metrics are stable across runs (accuracy: 0.9972 ± 0.0003). Data leakage was ruled out through strict temporal split ordering and pre-split SMOTE application exclusively on training folds. Confusion matrix and threshold-based analyses confirm strong classification performance with minimal false positives. SHAP-based explainability analysis further enhances the trustworthiness of the system by identifying the most influential multimodal predictors—past defect history, static analysis signals, and cyclomatic complexity—thereby contributing to transparent and accountable AI-driven infrastructure management. The presented framework advances the state of the art in LLM-driven multimodal systems by demonstrating how transformer intelligence, when applied to heterogeneous software artifact data streams, can enable reproducible, cloud-scalable, and interpretable maintenance pipelines for complex digital environments.
Software vulnerabilities remain a critical threat to modern software systems, while existing detection approaches often suffer from high computational cost, limited scalability, and severe class imbalance in real-world datasets. To address these challenges, this study proposes H-LLM-IM, an adaptive imbalance-aware hybrid framework for efficient software vulnerability detection. The proposed framework leverages semantic code embeddings extracted from a pre-trained code language model (CodeBERT) and integrates them with lightweight machine learning classifiers, thereby avoiding expensive fine-tuning of large language models.A key contribution of H-LLM-IM is an adaptive imbalance-aware learning mechanism that dynamically regulates imbalance mitigation intensity through controlled oversampling and adaptive reweighting based on minority-class performance feedback. Extensive experiments conducted on the Big-Vul benchmark dataset evaluate four classifiers (Logistic Regression, SVM, Random Forest, and XGBoost) under multiple imbalance-handling scenarios, including static and adaptive strategies.The results demonstrate that the proposed adaptive framework substantially improves minority-class vulnerability detection, achieving up to a 2.5-fold increase in F1-score and up to 83% improvement in MCC compared to the no-imbalance baseline. Importantly, these performance gains are obtained while maintaining practical training time and controlled memory growth. In particular, Logistic Regression and XGBoost exhibit the most favorable performance–efficiency trade-off, highlighting the scalability and practical applicability of H-LLM-IM for large-scale vulnerability analysis under severe class imbalance.
Fatma Altınsoy· Süleyman Demirel Üniversites...· 0 citations
In live-streaming e-commerce systems, massive volumes of user-generated danmaku constitute a critical yet underutilized source of business intelligence. However, transforming such unstructured, noisy, and highly context-dependent textual data into structured and actionable knowledge remains a fundamental challenge for enterprise information systems. To address this issue, this study proposes TaSC-LLM, an LLM-enabled topic recognition method for constructing interpretable topic measurements from unstructured user-generated content. The proposed framework integrates topic taxonomy construction and zero-shot classification into a unified semantic reasoning pipeline. Unlike conventional topic modeling or supervised classification approaches, TaSC-LLM leverages chain-of-thought reasoning, multi-stage taxonomy induction, sliding window context modeling, and self-consistency verification to eliminate reliance on predefined label spaces and annotated training data. This design allows the system to dynamically construct and update topic taxonomies while ensuring interpretability, robustness, and cross-scenario adaptability. Empirical evaluation on three large-scale live-streaming e-commerce danmaku datasets shows that TaSC-LLM achieves strong taxonomy coverage, classification accuracy, and agreement with expert annotations. The findings suggest that LLM-based reasoning can help convert unstructured user-generated text into interpretable topic measures for downstream empirical and managerial analysis. While the present evaluation is conducted offline, TaSC-LLM provides a methodological foundation for future business applications that can be further examined under multi-session, multi-platform, and deployment-oriented conditions.
This research proposes the first end-to-end multi-task architecture that jointly detects Python syntax errors and localizes their exact line position by adaptively fusing CodeBERT’s semantic embeddings with handcrafted syntactic features via a lightweight gating mechanism, without relying on Abstract Syntax Trees.
Rozali Ilham, Bahtiar Imran, Hasan Basri et al.· 0 citations
Detecting vulnerabilities in source code remains a major challenge as modern software systems increasingly span multiple programming languages. This work presents a deep learning approach for multilingual vulnerability detection that emphasizes semantic transfer rather than architectural complexity. The proposed method combines contextualized embeddings from CodeBERT/GraphCodeBERT with a BiLSTM and attention mechanism to capture code semantics and adopts a cross-lingual transfer setting where models trained on one language (e.g., Python) are evaluated on another (e.g., Java). To improve robustness under data imbalance, SMOTE and stratified cross-validation are incorporated into the training process. Experiments on Juliet and CodeXGLUE show that the model achieves an F1-score of about 0.87 and a ROC-AUC of 0.85 in intra-language settings, while guided fine-tuning improves cross-language F1-score by an average of 0.18 and ROC-AUC by approximately 0.13 compared with the direct transfer baseline. These results suggest that stabilizing semantic representations during transfer is key to improving generalization while maintaining practical efficiency under moderate computational constraints.
Tuan Nguyen Kim, Nin Ho Le Viet, Chieu Ta Quang· Engineering, Technology &...· 0 citations
The significance of this work lies in demonstrating that a deployable, explainable detector can be assembled from compact components, and an edge-type ablation study, a cross-dataset evaluation, and a per-vulnerability analysis are reported to characterize the approach.
Ahmed M. Elalfy, G. Ebrahim, M. B. Mansour· Computers· 0 citations
