SCICO

Kekül and Polatgi̇l compared ML models for classifying software quality attributes from user reviews of mobile health apps. The best configuration used TF-IDF vectorization, Random Forest classifier, and SMOTE oversampling, scoring an average F1 of 85%. authors.elsevier.com/a/1michc7X5IcB8

Leite et al. introduce DbC-GPT, an LLM that generates formal postcondition specifications for Solidity smart contracts. It was evaluated by measuring the precision of generated specifications across 3 standards using few-shot prompting and fine-tuning. authors.elsevier.com/a/1mhYYc7X5IcAm

Hyatt and Dewey developed a mutation-based end-to-end fuzzer for the Swift compiler that generates input programs. It takes seed Swift programs, mutates them with type-preserving transformations, and runs them to trigger compilation failures or crashes. doi.org/10.1016/j.scic…

Di Meglio et al. introduce E2E-Loader, generating web app performance test workloads from existing end-to-end GUI tests. A preliminary industrial evaluation showed that the generated workloads were comparable to those created manually. authors.elsevier.com/a/1mgmmc7X5IcAI

Benjamin Aziz introduces a formal, set-theoretic framework for defining security policies in database-defined networks with 3 models. These models can be enforced using various approaches, leading to different levels of expressivity. authors.elsevier.com/a/1mefSc7X5IcAg

Hashemi et al. developed JS-TOD, a tool that detects order-dependent flaky tests in Jest by systematically reordering test execution. JS-TOD was evaluated on 81 JavaScript projects and correctly reordered tests in 85% projects. doi.org/10.1016/j.scic…

Leger et al. introduce FASOW, a flexible agent-based simulator architecture for social network word-of-mouth modeling. FASOW was evaluated by replicating two previously published agent-based models and providing an online functional proof of concept. authors.elsevier.com/a/1mdkQc7X5Ic8t

Pérez et al. integrate multi-objective genetic algorithms with Metamorphic Testing and simulation into MT-EA4Cloud to optimize cloud performance and energy use. An empirical study shows that NSGA-II and higher mutation rates give the best results. doi.org/10.1016/j.scic…

Houichime and el Amrani introduce a dynamic, language-portable framework for call graph visualization and design pattern detection. It evaluates symbolic sequence properties and their impact on classification, validating the sequential representation. authors.elsevier.com/a/1mXOzc7X5Ic7j

Zhang et al. introduce PAL4JavaWeb, a generator of setup instructions for Java web apps based on static analysis and LLMs. It was evaluated on 35 projects and validated with open (and merged) pull requests in 13 GitHub projects. authors.elsevier.com/a/1mZW3c7X5Ic86

Chen et al. introduce DATVD, a vulnerability detection method that combines graph embeddings, dynamic attention, and hybrid convolutional pooling. It is evaluated across 3 public real-world datasets, showing improved accuracy compared to Devign. authors.elsevier.com/a/1mWTxc7X5Ic7G

Zhu et al. introduce CAMEL, a bug severity prediction framework that uses semantic feature fusion from CodeBERT and dynamic class weight optimization. It is evaluated on 3,342 method-level code snippets, showing significant improvements over baselines. authors.elsevier.com/a/1mU0Ac7X5Ic7S

Kangjian Zhou presents a five-year field diagnosis of cross-project defect prediction, centering on the ManualDown baseline. It evaluates recent models and finds that most achieve limited improvement over ManualDown, with LLMs as a tentative exception. authors.elsevier.com/a/1mV9sc7X5Ic8I

Aldawsari and Aldosari present an empirical study of security vulnerabilities in LLM-generated code, analyzing different models, languages, and prompts. Prompt engineering can substantially improve security, reducing vulnerabilities by up to 77%. authors.elsevier.com/a/1mW78c7X5Ic7k

Seyfipoor et al. introduce a CPS-oriented, urgency–based, adaptive real-time scheduling framework for advanced driver assistance systems. This was evaluated via software simulations, showing fewer deadline misses and context switches over baselines. authors.elsevier.com/a/1mXUac7X5Ic6I

Cataño presents a case study formalizing a chat system with the Event-B formal method and checking consistency via invariants and proof obligations. The study describes best practices for designing and verifying similar event-driven messaging systems. doi.org/10.1016/j.scic…

Rad and Khamespanah introduce an actor-based immune system framework for smart homes to detect anomalies in IoT sensor data and perform root cause analysis. The approach was implemented using an autoencoder, achieving 96% precision and a 98% F1 score. authors.elsevier.com/a/1mKAac7X5EoNY

Kassing and Giesl developed an annotated dependency pair framework for automatically proving almost-sure termination of probabilistic term rewrite systems, both for full and innermost rewriting. The framework is implemented in the AProVE tool. doi.org/10.1016/j.scic…

Gómez-Barrera et al. propose CFFitST for iterative fine-tuning sentence transformers to boost few-shot classification. It outperformed SetFit, achieving an F1 score of 84% on NLBSE 2024 competition (multi-class issue classification). authors.elsevier.com/a/1mG-bc7X5EoLl

Azhir et al. introduce QualCode, a data-driven framework that predicts software maintainability using ISO/IEC 25010 sub-characteristics. It showed better accuracy than existing tools in maintainability assessment after an evaluation on Java projects. authors.elsevier.com/a/1mBTec7X5EoIi