IET Software最新文献

In today’s rapidly changing software development landscape, ensuring code quality is essential to reliability, maintainability, and security among other aspects. Identifying code quality issues can be tackled; however, implementing code quality improvements can be a complex and time-consuming task. To address this problem, we present a novel methodology designed to assist developers by suggesting alternative code snippets that not only match the functionality of the original code but also improve its quality based on predefined metrics. Our system is based on a language-agnostic approach that allows the analysis of code snippets written in different programming languages. It employs advanced techniques to assess functional similarity and evaluates syntactic similarity, suggesting alternatives that minimize the need for extensive modification. The evaluation of our system on multiple axes demonstrates the effectiveness of our approach in providing usable code alternatives that are both functionally equivalent and syntactically similar to the original snippets, while significantly improving quality metrics. We argue that our methodology and tool can be valuable for the software engineering community, bridging the gap between the identification of code quality problems and the implementation of practical solutions that improve software quality.

Commit classification is an important task in software maintenance, since it helps software developers classify code changes into different types according to their nature and purpose. This allows them to better understand how their development efforts are progressing, identify areas where they need improvement, and make informed decisions about when and how to release new versions of their software. However, existing methods are all discriminative models, usually with complex architectures that require additional output layers to produce class label probabilities, making them task-specific and unable to learn features across different tasks. Moreover, they require a large amount of labeled data for fine tuning, and it is difficult to learn effective classification boundaries in the case of limited labeled data. To solve the above problems, we propose a generative framework that incorporates prompt tuning for commit classification with external knowledge (IPCK), which simplifies the model structure and learns features across different tasks, only based on the commit message information as the input. First, we proposed a generative framework based on T5 (text-to-text transfer transformer). This encoder–decoder construction method unifies different commit classification tasks into a text-to-text problem, simplifying the model’s structure by not requiring an extra output layer. Second, instead of fine tuning, we design a prompt tuning solution that can be adopted in few-shot scenarios with only limited samples. Furthermore, we incorporate external knowledge via an external knowledge graph to map the probabilities of words into the final labels in the speech machine step to improve performance in few-shot scenarios. Extensive experiments on two open available datasets demonstrate that our framework can solve the commit classification problem simply but effectively for both single-label binary classification and single-label multiclass classification purposes with 90% and 83% accuracy. Further, in the few-shot scenarios, our method improves the adaptability of the model without requiring a large number of training samples for fine tuning.

With the development of intelligent manufacturing and the wider application of the Internet of Things (IoT), it is crucial to fuse heterogeneous sensor data from multiple sources. However, the current data fusion methods still have problems, such as low accuracy of fused data, insufficient data integrity, poor data fusion efficiency, and poor scalability of fusion methods. In response to these issues, this article explores a multisource heterogeneous data fusion method based on the Prophet algorithm digital twin drive to improve the fusion effect of sensor data and provide more support for subsequent decision-making. The article first used curve and sequence alignment to extract data features and then analyzed the trend of data changes using the Prophet algorithm. Afterward, this article constructed a digital twin model to provide analytical views and data services. In conclusion, this paper used tensor decomposition to merge text and image data from sensor data. Deep learning algorithms and Kalman filtering techniques were also examined to confirm the efficacy of data fusion under the Prophet algorithm. The experimental results showed that after fusing the data using the Prophet algorithm, the average accuracy can reach 92.63%, while the average resource utilization at this time was only 9.97%. The results showed that combining Prophet with digital twin technology can achieve higher accuracy, fusion efficiency, and better scalability. The research in this paper can provide new ideas and means for the fusion and analysis of heterogeneous data from multiple sources.

Context: Graphical user interface (GUI) testing of mobile applications (apps) is significant from a user perspective to ensure that the apps are visually appealing and user-friendly. Pattern-based GUI testing (PBGT) is an innovative model-based testing (MBT) approach designed to enhance user satisfaction and reusability while minimizing the effort required to model and test UIs of mobile apps. In the literature, several primary studies have been conducted in the domain of PBGT.

Problem: The current state-of-the-art lacks comprehensive secondary studies within the PBGT domain. To our knowledge, this area has insufficient focus on in-depth research. Consequently, numerous challenges and limitations persist in the existing literature.

Objective: This study aims to fill the gaps mentioned above in the existing body of knowledge. We highlight popular research topics and analyze their relationships. We explore current state-of-the-art approaches and techniques, a taxonomy of tools and modeling languages, a list of reported UI test patterns (UITPs), and a taxonomy of writing UITPs. We also highlight practical challenges, limitations, and gaps in the targeted research area. Furthermore, the current study intends to highlight future research directions in this domain.

Method: We conducted a systematic literature review (SLR) on PBGT in the context of Android and web apps. A hybrid methodology that combines the Kitchenham and PRISMA guidelines is adopted to achieve the targeted research objectives (ROs). We perform a keyword-based search on well-known databases and select 30 (out of 557) studies.

Results: The current study identifies 11 tools used in PBGT and devises a taxonomy to categorize these tools. A taxonomy for writing UITPs has also been developed. In addition, we outline the limitations of the targeted research domain and future directions.

Conclusion: This study benefits the community and readers by better understanding the targeted research area. A comprehensive knowledge of existing tools, techniques, and methodologies is helpful for practitioners. Moreover, the identified limitations, gaps, emerging trends, and future research directions will benefit researchers who intend to work further in future research.

In software engineering, selecting the appropriate architectural style for software systems is risky and sensitive. The selection process is a multicriteria decision-making (MCDM) problem. Consequently, selecting a suitable architecture is a key challenge in software development. This study presents an automated hybrid methodology based on the analytic hierarchy process (AHP) and fuzzy analytic hierarchy process (FAHP) to evaluate and suggest multiple architectural styles based on quality attributes (QAs) alone rather than relying on expert opinions. A Tera-PROMISE dataset is presented to illustrate the proposed methodology and then compare the result of the methodology with expert judgments. Moreover, to support the proposed methodology, a case study is carried out to compare the proposed method to previous studies.

With the development of artificial intelligence and blockchain technology, the training of deep learning models needs large computing resources. Meanwhile, the Proof of Work (PoW) consensus mechanism in blockchain systems often leads to the wastage of computing resources. This article combines distributed deep learning (DDL) with blockchain technology and proposes a blockchain consensus scheme based on the proof of distributed deep learning work (BCDDL) to reduce the waste of computing resources in blockchain. BCDDL treats DDL training as a mining task and allocates different training data to different nodes based on their computing power to improve the utilization rate of computing resources. In order to balance the demand and supply of computing resources and incentivize nodes to participate in training tasks and consensus, a dynamic incentive mechanism based on task size and computing resources (DIM-TSCR) is proposed. In addition, in order to reduce the impact of malicious nodes on the accuracy of the global model, a model aggregation algorithm based on training data size and model accuracy (MAA-TM) is designed. Experiments demonstrate that BCDDL can significantly increase the utilization rate of computing resources and diminish the impact of malicious nodes on the accuracy of the global model.

Context: As more time has been spent on code comprehension activities during software development, automatic code summarization has received much attention in software engineering research, with the goal of enhancing software comprehensibility. In the meantime, it is prevalently known that a good knowledge about the declaration and the use of method parameters can effectively enhance the understanding of the associated methods. A traditional approach used in software development is to declare the types of method parameters.

Objective: In this work, we advocate parameter-level code summarization and propose a novel approach to automatically generate parameter summaries of a given method. Parameter summarization is considerably challenging, as neither do we know the kind of information of the parameters that can be employed for summarization nor do we know the methods for retrieving such information.

Method: We present paramTrans, which is a novel approach for parameter summarization. paramTrans characterizes the semantic features from parameter-related information based on transformer; it also explores three fusion strategies for absorbing the method-level information to enhance the performance. Moreover, to retrieve parameter-related information, a parameter slicing algorithm (named paramSlice) is proposed, which slices the parameter-related node from the abstract syntax tree (AST) at the statement level.

Results: We conducted experiments to verify the effectiveness of our approach. Experimental results show that our approach possesses an effective ability in summarizing parameters; such ability can be further enhanced by understanding the available summaries about individual methods, through the introduction of three fusion strategies.

Conclusion: We recommend developers employ our approach as well as the fusion strategies to produce parameter summaries to enhance the comprehensibility of code.

Web-based applications are popular in demand and usage. To facilitate the development of web-based applications, the software engineering community developed multiple web application frameworks, one of which is Flask. Flask is a popular web framework that allows developers to speed up and scale the development of web applications. A review of the software engineering literature revealed that the Stack Overflow (SO) website has proven its effectiveness in providing a better understanding of multiple subjects within the software engineering field. This study aims to analyze SO Flask-related questions to gain a better understanding of the stance of Flask on the website. We identified a set of 70,230 Flask-related questions that we further analyzed to estimate how the interest towards the framework evolved over time on the website. Afterward, we utilized the Latent Dirichlet Allocation (LDA) algorithm to identify Flask-related topics that are discussed within the set of the identified questions. Moreover, we leveraged a number of proxy measures to examine the difficulty and popularity of the identified topics. The study found that the interest towards Flask has been generally increasing on the website, with a peak in 2020 and drops in the following years. Moreover, Flask-related questions on SO revolve around 12 topics, where Application Programming Interface (API) can be considered the most popular topic and background tasks can be considered the most difficult one. Software engineering researchers, practitioners, educators, and Flask contributors may find this study useful in guiding their future Flask-related endeavors.

The technique of software defect prediction aims to assess and predict potential defects in software projects and has made significant progress in recent years within software development. In previous studies, this technique largely relied on supervised learning methods, requiring a substantial amount of labeled historical defect data to train the models. However, obtaining these labeled data often demands significant time and resources. In contrast, software defect prediction based on unsupervised learning does not depend on known labeled data, eliminating the need for large-scale data labeling, thereby saving considerable time and resources while providing a more flexible solution for ensuring software quality. This paper conducts software defect prediction using unsupervised learning methods on data from 16 projects across two public datasets (PROMISE and NASA). During the feature selection step, a chi-squared sparse feature selection method is proposed. This feature selection strategy combines chi-squared tests with sparse principal component analysis (SPCA). Specifically, the chi-squared test is first used to filter out the most statistically significant features, and then the SPCA is applied to reduce the dimensionality of these significant features. In the clustering step, the dot product matrix and Pearson correlation coefficient (PCC) matrix are used to construct weighted adjacency matrices, and a clustering overlap method is proposed. This method integrates spectral clustering, Newman clustering, fluid clustering, and Clauset–Newman–Moore (CNM) clustering through ensemble learning. Experimental results indicate that, in the absence of labeled data, using the chi-squared sparse method for feature selection demonstrates superior performance, and the proposed clustering overlap method outperforms or is comparable to the effectiveness of the four baseline clustering methods.

Software defect prediction (SDP) is a crucial phase preceding the launch of software products. Cross-project defect prediction (CPDP) is introduced for the anticipation of defects in novel projects lacking defect labels. CPDP can use defect information of mature projects to speed up defect prediction for new projects. So that developers can quickly get the defect information of the new project, so that they can test the software project pertinently. At present, the predominant approaches in CPDP rely on deep learning, and the performance of the ultimate model is notably affected by the quality of the training dataset. However, the dataset of CPDP not only has few samples but also has almost no label information in new projects, which makes the general deep-learning-based CPDP model not ideal. In addition, most of the current CPDP models do not fully consider the enrichment of classification boundary samples after cross-domain, leading to suboptimal predictive capabilities of the model. To overcome these obstacles, we present contrastive learning pretraining for CPDP (ConCPDP), a CPDP method integrating contrastive pretraining and category boundary adjustment. We first perform data augmentation on the source and target domain code files and then extract the enhanced data as an abstract syntax tree (AST). The AST is then transformed into an integer sequence using specific mapping rules, serving as input for the subsequent neural network. A neural network based on bidirectional long short-term memory (Bi-LSTM) will receive an integer sequence and output a feature vector. Then, the feature vectors are input into the contrastive module to optimise the feature extraction network. The pretrained feature extractor can be fine-tuned by the maximum mean discrepancy (MMD) between the feature distribution of the source domain and the target domain and the binary classification loss on the source domain. This paper conducts a large number of experiments on the PROMISE dataset, which is commonly used for CPDP, to validate ConCPDP’s efficacy, achieving superior results in terms of F₁ measure, area under curve (AUC), and Matthew’s correlation coefficient (MCC).