arXiv subjects

Elvys Soares

Publications and source records attributed to Elvys Soares.

6 recordsLinked to original sources

arXiv2026

An Empirical Study of Gemini 3 for Detecting Natural Language Test Smells in Manual Test Cases

Manual testing, in which testers follow natural language instructions to validate system behavior, remains essential for uncovering issues that are difficult to capture with automation. However, manual test cases often contain test smells, quality issues such as ambiguity, redundancy, or missing checks that reduce reliability, maintainability, and reproducibility. Existing detection approaches largely depend on manually engineered rules and thus struggle to generalize and scale across heterogeneous test suites. In our previous work, we assessed the feasibility of using Small Language Models (SLMs) for test smell detection by evaluating GEMMA-3-4B, LLAMA-3.2-3B, and PHI-4-14B on test steps from 143 real-world Ubuntu test cases, covering seven smell types. PHI-4-14B achieved the best performance. In this article, we investigate whether a contemporary Large Language Model (GEMINI-3-PRO-PREVIEW) available at the time of the study can identify test smells in natural language manual test cases using a prompt-based, whole-test-case analysis strategy. Unlike approaches that analyze individual test steps in isolation, our approach evaluates complete test cases, enabling the model to consider relationships and dependencies among test steps. We evaluate the approach on 100 Ubuntu test cases covering seven test smell types and compare its performance against previously evaluated SLMs, including GEMMA-3-4B, LLAMA-3.2-3B, and PHI-4-14B. Our results show that GEMINI-3-PRO-PREVIEW outperforms the SLMs, while producing actionable explanations that can help practitioners revise manual test cases for greater clarity and consistency. We also find that test smells are pervasive in practice, with nearly one detected test smell per step on average, highlighting the need for scalable and automated quality support for manual testing artifacts.

cs.SE↗

arXiv2025

Agentic LMs: Hunting Down Test Smells

Test smells reduce test suite reliability and complicate maintenance. While many methods detect test smells, few support automated removal, and most rely on static analysis or machine learning. This study evaluates models with relatively small parameter counts - Llama-3.2-3B, Gemma-2-9B, DeepSeek-R1-14B, and Phi-4-14B - for their ability to detect and refactor test smells using agent-based workflows. We assess workflows with one, two, and four agents over 150 instances of 5 common smells from real-world Java projects. Our approach generalizes to Python, Golang, and JavaScript. All models detected nearly all instances, with Phi-4-14B achieving the best refactoring accuracy (pass@5 of 75.3%). Phi-4-14B with four-agents performed within 5% of proprietary LLMs (single-agent). Multi-agent setups outperformed single-agent ones in three of five smell types, though for Assertion Roulette, one agent sufficed. We submitted pull requests with Phi-4-14B-generated code to open-source projects and six were merged.

cs.SE↗

arXiv2025

Investigating the Performance of Small Language Models in Detecting Test Smells in Manual Test Cases

Manual testing, in which testers follow natural language instructions to validate system behavior, remains crucial for uncovering issues not easily captured by automation. However, these test cases often suffer from test smells, quality issues such as ambiguity, redundancy, or missing checks that reduce test reliability and maintainability. While detection tools exist, they typically require manual rule definition and lack scalability. This study investigates the potential of Small Language Models (SLMs) for automatically detecting test smells. We evaluate Gemma3, Llama3.2, and Phi-4 on 143 real-world Ubuntu test cases, covering seven types of test smells. Phi-4 achieved the best results, reaching a pass@2 of 97% in detecting sentences with test smells, while Gemma3 and Llama3.2 reached approximately 91%. Beyond detection, SLMs autonomously explained issues and suggested improvements, even without explicit prompt instructions. They enabled low-cost, concept-driven identification of diverse test smells without relying on extensive rule definitions or syntactic analysis. These findings highlight the potential of SLMs as efficient tools that preserve data privacy and can improve test quality in real-world scenarios.

cs.SE↗

arXiv2024

Evaluating Large Language Models in Detecting Test Smells

Test smells are coding issues that typically arise from inadequate practices, a lack of knowledge about effective testing, or deadline pressures to complete projects. The presence of test smells can negatively impact the maintainability and reliability of software. While there are tools that use advanced static analysis or machine learning techniques to detect test smells, these tools often require effort to be used. This study aims to evaluate the capability of Large Language Models (LLMs) in automatically detecting test smells. We evaluated ChatGPT-4, Mistral Large, and Gemini Advanced using 30 types of test smells across codebases in seven different programming languages collected from the literature. ChatGPT-4 identified 21 types of test smells. Gemini Advanced identified 17 types, while Mistral Large detected 15 types of test smells. Conclusion: The LLMs demonstrated potential as a valuable tool in identifying test smells.

cs.SE↗

arXiv2024

A Catalog of Transformations to Remove Smells From Natural Language Tests

Test smells can pose difficulties during testing activities, such as poor maintainability, non-deterministic behavior, and incomplete verification. Existing research has extensively addressed test smells in automated software tests but little attention has been given to smells in natural language tests. While some research has identified and catalogued such smells, there is a lack of systematic approaches for their removal. Consequently, there is also a lack of tools to automatically identify and remove natural language test smells. This paper introduces a catalog of transformations designed to remove seven natural language test smells and a companion tool implemented using Natural Language Processing (NLP) techniques. Our work aims to enhance the quality and reliability of natural language tests during software development. The research employs a two-fold empirical strategy to evaluate its contributions. First, a survey involving 15 software testing professionals assesses the acceptance and usefulness of the catalog's transformations. Second, an empirical study evaluates our tool to remove natural language test smells by analyzing a sample of real-practice tests from the Ubuntu OS. The results indicate that software testing professionals find the transformations valuable. Additionally, the automated tool demonstrates a good level of precision, as evidenced by a F-Measure rate of 83.70%

cs.SE↗

arXiv2023

Manual Tests Do Smell! Cataloging and Identifying Natural Language Test Smells

Background: Test smells indicate potential problems in the design and implementation of automated software tests that may negatively impact test code maintainability, coverage, and reliability. When poorly described, manual tests written in natural language may suffer from related problems, which enable their analysis from the point of view of test smells. Despite the possible prejudice to manually tested software products, little is known about test smells in manual tests, which results in many open questions regarding their types, frequency, and harm to tests written in natural language. Aims: Therefore, this study aims to contribute to a catalog of test smells for manual tests. Method: We perform a two-fold empirical strategy. First, an exploratory study in manual tests of three systems: the Ubuntu Operational System, the Brazilian Electronic Voting Machine, and the User Interface of a large smartphone manufacturer. We use our findings to propose a catalog of eight test smells and identification rules based on syntactical and morphological text analysis, validating our catalog with 24 in-company test engineers. Second, using our proposals, we create a tool based on Natural Language Processing (NLP) to analyze the subject systems' tests, validating the results. Results: We observed the occurrence of eight test smells. A survey of 24 in-company test professionals showed that 80.7% agreed with our catalog definitions and examples. Our NLP-based tool achieved a precision of 92%, recall of 95%, and f-measure of 93.5%, and its execution evidenced 13,169 occurrences of our cataloged test smells in the analyzed systems. Conclusion: We contribute with a catalog of natural language test smells and novel detection strategies that better explore the capabilities of current NLP mechanisms with promising results and reduced effort to analyze tests written in different idioms.

cs.SE↗