Traceability links between requirements and source code are crucial for maintaining software quality, yet traditional traceability methods exhibit critical limitations. Current methods predominantly rely on semantic similarity among artifacts, failing to account for link transitivity and the underlying rationale governing traceability link formation. To address these challenges, this study introduces HGTRTracer, a novel framework that integrates Large Language Models (LLMs) with Graph Neural Networks (GNNs) to advance traceability analysis. The proposed method innovatively incorporates attribution reasoning via LLMs to derive explicit “Reason nodes” that justify traceability links, with a Heterogeneous Graph Transformer (HGT) captures structural dependencies among nodes and edges to improve link prediction accuracy. Experiments conducted on 10 open-source projects demonstrate that HGTRTracer outperforms the state-of-the-art baseline (HGNNLink) by 12.90% in F1-score on average. Beyond performance gains, the framework enhances interpretability through its attribution-aware design, providing actionable insights for traceability decisions.

Current research on requirements-to-code traceability predominantly focuses on improving the performance of algorithms or models, while neglecting the quality of project artifacts themselves.Nevertheless, high-quality data can significantly improve the performance of a model, and disregarding data quality can be challenging to rectify even with sophisticated models.Often, high-quality data contributes substantially more to model performance enhancement than improvements to the model's structure itself.In real-world software development, requirements and code artifacts are the primary data sources for establishing traceability links.However, there is a lack of effective guidance on what factors of requirements and code artifacts are conducive to traceability.To address this issue, this study proposes five metrics that are verified to have a close impact on the requirementsto-code traceability: cyclomatic complexity, annotation ratio, co-occurrence ratio, noun-verb ratio, and noise ratio.Based on a systematic mapping study, we selected 11 open-source projects from 20 relevant publications to conduct experiments, and employed Spearman correlation analysis to validate the association between the proposed metrics and traceability.The experimental results show that the proposed five metrics can significantly affect the traceability of requirements-to-code, of which the cyclomatic complexity and the annotation ratio are more influential.

Abstract Named data networking (NDN) is considered a novel architecture of the next-generation Internet that delivers content by names. However, the human-readable name may potentially leak users’ privacy. Existing solutions have focused on encrypting to protect privacy, but they are neither efficient in the case of one publisher and multiple subscribers nor successful in supporting prefix matching. To address the above challenges, we propose an efficient and flexible name scheme with privacy preservation, which combines message-driven encryption with Bloom Filter. Firstly, message-driven encryption is utilized to protect name privacy, thus supporting efficient and secure encrypted name matching. Secondly, each name is divided into multiple components, and then each component is encrypted separately to support flexible prefix matching. Lastly, the technology of Bloom Filter with random numbers is explored to improve the efficiency and accuracy of name matching. Security and performance analysis shows that the proposed scheme effectively enhances the efficiency and accuracy of data matching while protecting name privacy.

In the field of software traceability (ST), machine learning (ML) has become a common and effective method for automated issue-commit link recovery. The features extracted from issue and commit artifacts are composed of significantly different types of data, such as issue summaries, diff codes, and hashes. Such complex and diverse data poses a challenge to conventional ML methods. To overcome this challenge, we propose a novel model named FRELinker, which trains independent classifiers based on the type of issue-commit training data, fully leveraging the effectiveness of ML for single type data, and then fuses the classifiers. Specifically, we categorize the features into four types: textual features, code features, non-textual features, and similarity features, and extend text similarity features by adding hybrid textual similarity measures. And then, we use a ranking method to select the optimal classifiers corresponding to these four types of features. Among them, the optimal classifier for textual features is Gradient Boosting (GB), the optimal classifier for code features is Logistic Regression (LR), and the optimal classifier for non-textual features and similarity features is Random Forest (RF). Finally, we use a Bayesian optimization model to fuse these four classifiers. Experimental results show that our method outperforms competing methods Hybrid-Linker and DeepLink in terms of Precision, Recall, and F-measure on six real-world open-source software (OSS) datasets, demonstrating significant performance advantages in complex and diverse data.

In the task of requirements-to-code traceability link recovery, the continues growth of software scale has led to diminishing differences between indices and more complex nonlinear relationships within the data. This results in the performance decline of the most widely used information retrieval methods and machine learning methods when handling this task. Therefore, we propose a requirement traceability method based on heterogeneous graph attention networks, named HANTracer. The model integrates the high-dimensional vectors generated by a pre-trained model as node features to deepen the dif-ferentiation between nodes and enhances node representations with contextual information learned from the graph structure. Additionally, it utilizes the properties of heterogeneous edges to construct various edge features, such as code calling, code inheritance, and text similarity relationships, aiding the model in understanding and utilizing the relationships between different types of data elements. By incorporating average pooling layers and multiple fully connected layers, the HANTracer model is further improved to enhance its ability to extract nonlinear features. Experimental results indicate that HANTracer achieves an average F1 performance higher than the state-of-the-art methods TAROT by 100.30% and DF4RT by 46.27% on seven real-world open source software (OSS) datasets, demonstrating significant performance advantages in large-scale and complex data environments.

In recent years, increasing machine learning has been applied to Requirements Traceability Recovery (RTR). The performance of these approaches is not satisfactory. Besides, most of them perform well in recovering traceability links between specific artifacts but cannot maintain consistent performance in different scenarios. To alleviate this problem, we propose a novel approach based on Multi-Layered Gradient Boosting Decision Trees (mGBDT) for RTR which is called MLTracer. MLTracer stacks several GBDTs and learns hierarchical representations of artifact link features. Through layer-by-layer training, it adapts to the feature distribution in different scenarios to improve generalization ability. MLTracer is evaluated on five software projects and compared with six state-of-the-art approaches. The results indicate that MLTracer achieves an average F1 score of 0.6153, outperforming six baseline approaches across all datasets. It proves that MLTracer achieves state-of-the-art performance and has strong applicability in actual industry.

Traceability refers to the potential for traces to be established (i.e., created and maintained) and used. Bug Traceability (BT) is critical for enhancing software quality, reducing maintenance costs, and boosting team efficiency. To explore the trends and advancements of BT, we conduct a systematic mapping study (SMS). We initially retrieve 4674 citations from 7 databases spanning 2014 to 2023, and 24 primary studies meet the rigorous selection criteria. Our study identifies 6 types of bug trace links, 8 traceability strategies, and 47 bug traceability recovery (BTR) techniques. Among them, 47 BTR techniques can be further classified into 6 categories. At the same time, we perform statistics on 113 datasets and 16 evaluation metrics used to assess the performance of BTR techniques proposed in the primary studies. In evaluating the overall quality of the primary studies, 8 dimensions are utilized to support technology transfer, categorizing the overall quality into 4 levels: poor, middle, good, and excellent, with 79% of primary studies evaluated at a good level. This study not only furnishes a clear definition of BT for scholarly reference, but also highlights that information retrieval (IR), machine learning (ML) and deep learning (DL) techniques are the mainstream techniques used for BTR.

Information Retrieval (IR), Machine Learning (ML), and Deep Learning (DL) have become mainstream methods for traceability link recovery. However, IR-based methods face the challenge of low precision, while DL-based methods require large-scale training data to achieve better performance. In this paper, we propose a novel model XWCoDe, which apply XGBoost combined with a weighted code dependency strategy to traceability link recovery domain. In order to refine the initial candidate links generated by the XGBoost model, the strategy only modifies low confidence candidate links and pioneers the use of graph embedding technology node2vec to calculate the importance of each code dependency relationship. The experimental results show that the average F1 score of XW CoDe on 4 datasets and 9 training/testing ratios is 12.93 % higher than the state-of-the-art method DF4RT.

In software stores, user reviews are the most direct feedback on product experiences and requirements. Sentiment analysis (SA) technology has been proven to be efficient for review mining-based requirements elicitation. Our objective is to identify the trend and actuality in SA for requirements elicitation from app reviews. We conduct a systematic mapping study, retrieving 740 citations from 2013 to 2023, and 33 articles are retained as primary studies. The overall research posture is increasing yearly. There are 24 self-crawled and seven specially created datasets. SVM is the most commonly used classifier, while BERT represents the most innovative algorithm with a good performance. Precision, recall, and F -measure are the most popular metrics. An SA framework based on feature granularity is proposed to guide newcomers in this field. It is proven that SA can effectively assist in transitioning from traditional methods to review mining in requirements elicitation. Overall, SA for requirements elicitation from app reviews is becoming an increasingly mature cross-research field. Future research can focus on building highly generalized and multilingual datasets, using more fine-grained feature extraction methods to improve requirements elicitation accuracy, and exploring BERT families or other attention-based deep learning models.

Machine learning (ML) has been widely used in trace link recovery (TLR) to reduce the manual maintenance cost of trace links by developers. However, the imbalanced distribution of valid links and invalid links seriously affects the performance of classifiers. Although a few studies have applied data balancing techniques (DBT) to ML-based TLR, none of them has systematically analyzed more effective combinations of them. Therefore, we perform an empirical study on three groups of control experiments to explore the impact of the combination of different ML methods with and without DBT on TLR efficiency. We compare the performance of supervised ML-based TLR and unsupervised ML-based TLR with and without DBT respectively. Then, we analyze the performance of the ensemble learning model (EM) with DBT on TLR. The experimental results on the 7 imbalance datasets of CoEST indicate that DBT has a positive effect on ML-based TLR. Specifically, the recall of the LR model increased by 0.5517 after combining with most DBTs on EasyClinic(ID-TC), while Tomek-link significantly improves the precision of K-Nearest Neighbor (KNN), Decision Tree (DT), LR, Support Vector Machine (SVM). The precision of LR increased from 0.5036 to 1.0. BalanceRF is best at increasing recall, reaching 1.0 on 4 datasets. Moreover,the improvement degree of ML-based TLR with DBT shows differences in terms of the size of datasets and the proportion of valid links.

Requirements traceability (RT) is crucial for requirement management and impact analysis of requirement change in software development.The applications of machine learning (ML) technologies to RT have received much attention.In this paper, we aim to provide the state-of-the-art progress of the studies on the intersection of ML and RT.A systematic mapping study (SMS) is conducted and 26 studies have been identified as primary studies.The results present 32 ML technologies and 7 enhancement strategies for establishing trace links.Besides, 46 datasets are utilized for validating the performance of these ML technologies.Additionally, the overall quality of these primary studies is at a good level.This study indicates that numerous studies have proved the potential of utilizing ML technologies for predicting emerging trace links in RT by utilizing existing traceability information.Moreover, open-source datasets are the most popular, which greatly improves the reproducibility of studies.However, there is still a gap between academia and industrial application because of the lack of industrial practice and guidance from practitioners.

Nowadays, many supervised learning techniques have been applied to requirements traceability recovery (RTR). However, the performance of these supervised learning techniques is still far from satisfactory, and exploring a more effective model is necessary. This paper proposes a new deep forest model for RTR(DF4RT) with a novel composition to improve the model's performance. The proposed model incorporates three feature representation methods, which not only information retrieval and query quality but also add distance. The DF4RT model is evaluated on four open-source projects and compared with nine state-of-the-art tracing approaches. The experimental results show that DF4RT improves precision by 94%, recall by 58%, and F-measure by 72% on average. We also conduct ablation experiments to explore the impact of the different features. It is the first time that deep forest is employed in requirements traceability. Our approach is effective for RTR with good interpretability, few parameters, and good performance in small-scale data.

Context: Software traceability (ST) refers to capturing associations in various artifacts. A growing interest has been in applying machine learning (ML) techniques to ST. Objective: The purpose of this work is to present a comprehensive review of the state-of-the-art progress on the intersection of ML and ST. Method: A systematic mapping study (SMS) is conducted. A total of 965 citations are retrieved from 2013 to 2022, among which 37 studies are selected as primary studies. Result: 32 ML technologies and 9 enhancement strategies for generating trace links have been identified. Besides, 90 datasets and 16 measures have been summarized, which are applied to evaluate the efficacy of the ML-based tracing techniques. The overall reproducibility of these primary studies is at a medium level. Conclusion: We have found that ML is playing a positive role in improving the accuracy and efficiency of ST. However, there are still some challenges such as reproducibility. Hence, researchers are suggested to pay more attention to standardization to improve the reproducibility of studies.

Software defect prediction technology has been widely used in improving the quality of software system. Most real software defect datasets tend to have fewer defective modules than defective-free modules. Highly class-imbalanced data typically make accurate predictions difficult. The imbalanced nature of software defect datasets makes the prediction model classifying a defective module as a defective-free one easily. As there exists the similarity during the different software modules, one module can be represented by the sparse representation coefficients over the pre-defined dictionary which consists of historical software defect datasets. In this study, we make use of dictionary learning method to predict software defect. We optimize the classifier parameters and the dictionary atoms iteratively, to ensure that the extracted features (sparse representation) are optimal for the trained classifier. We prove the optimal condition of the elastic net which is used to solve the sparse coding coefficients and the regularity of the elastic net solution. Due to the reason that the misclassification of defective modules generally incurs much higher cost risk than the misclassification of defective-free ones, we take the different misclassification costs into account, increasing the punishment on misclassification defective modules in the procedure of dictionary learning, making the classification inclining to classify a module as a defective one. Thus, we propose a cost-sensitive software defect prediction method using dictionary learning (CSDL). Experimental results on the 10 class-imbalance datasets of NASA show that our method is more effective than several typical state-of-the-art defect prediction methods.

Summary Thesaurus‐based, code‐related, and software‐specific query expansion techniques are the main contributions in free‐form query search. However, these techniques still could not put the most relevant query result in the first position because they lack the ability to infer the expansion words that represent the user needs based on a given query. In this paper, we discover that code changes can imply what users want and propose a novel query expansion technique with code changes (QECC). It exploits (changes, contexts) pairs from changed methods. On the basis of statistical learning from pairs, it can infer code changes for a given query. In this way, it expands a query with code changes and recommends the query results that meet actual needs perfectly. In addition, we implement InstaRec to perform QECC and evaluate it with 195 039 change commits from GitHub and our code tracker. The results show that QECC can improve the precision of 3 code search algorithms (ie, IR, Portfolio, and VF) by up to 52% to 62% and outperform the state‐of‐the‐art query expansion techniques (ie, query expansion based on crowd knowledge and CodeHow) by 13% to 16% when the top 1 result is inspected.

To enable the cod sarch results to run immediately without any subsequent modification, an intent-enhanced code search approach (IECS) is proposed.It has the ability of intent predicting to guess what else a user might do after obtaining the search results.Based on the intent-relevant semantic and structural matches, IECS improves the performance of code search by incorporating the intent for expansion.To perform IECS, the code search tool SnippetGen is implemented.Compared with CodeHow and Google Code Search (CS), SnippetGen outperforms them by 28.5% with a precision score of 0.846 (i.e., 84.6% of the first results are relevant).

With the popularization of software version control system and defect tracking tools, large amounts of software development data is recorded.How to effectively use these data to improve the quality of software development, has become a hot topic in recent years.Software defect prediction technology can take full advantage of the historical data to build predictive models and automatically detect defective modules for efficient software test to improve the quality of a software system.But the class-imbalanced data makes the prediction model classifying a modules as a defective-free one easily, while the misclassification of defective modules generally incurs much higher cost risk than the misclassification of defective-free ones.To resolve this problem, we propose a cost-sensitive software defect prediction method using dictionary learning.It iteratively optimizes the classifier parameters and the dictionary atoms, to ensure that the extracted features (sparse representation) are optimal for the trained classifier; Moreover, we take the different misclassification costs into account, increasing the punishment on misclassification defective modules in the procedure of dictionary learning, making classification inclining to classify a module as a defective one.Experimental results on the 10 classimbalanced data sets of NASA show that our method is more effective than other methods.

Changes in operating environment may result in the performance degradation to a SaaS software.Analyzing running log is an efficient method to locate this problem.However, as a long-running software, SaaS may generate huge log data which is difficult to analyze, and it lacks a systematic approach to implement management of the running log.These all threaten the timeliness of SaaS performance analysis.In this paper, we define a log format to standardize multi-source heterogeneous log data and construct a log model to support SaaS software performance analysis, where the two performance metrics of average response time and request timeout rate in the model are calculated by statistical measurement.Furthermore, a log management framework is given to support real-time big log data collection, access, calculation, storage and service, and the technology implementation of the framework is also given.Finally, a case study is given to illustrate and validate the effectiveness of the approach.

To make the code search (CS) become more effective, a novel query expansion with intents (QEI) is proposed, in which the intent refers to the common subsequent modifications of the search results. The intent is extracted from the modification history. Within the intent scope, the CS is speeded up based on the semantic and structural matches. The precision of the search results is also increased by expanding the query with the intent. Compared with CodeHow and Google CS, QEI outperforms them by 28.5% with a precision score of 0.846. (i.e. 84.6% of the first results are accepted directly by users).

Software defect prediction generally builds models from intra-project data.Lack of training data at the early stage of software testing limits the efficiency of prediction in practice.Thereby researchers proposed cross-project defect prediction using the data from other projects.Most previous efforts assumed the cross-project defect data have the same metrics set which means the metrics used and size of metrics set are same in the data of projects.However, in real scenarios, this assumption may not hold.In addition, software defect datasets have the class imbalance problem increasing the difficulty for the learner to predict defects.In this paper, we advance canonical correlation analysis for deriving a joint feature space for associating crossproject data and propose a novel support vector machine algorithm which incorporates the correlation transfer information into classifier design for cross-project prediction.Moreover, we take different misclassification costs into consideration to make the classification inclining to classify a module as a defective one, alleviating the impact of imbalanced data.Experiments on public heterogeneous datasets from different projects show that our method is more effective, compared to state-of-the-art methods.

Cross-project defect prediction trains a prediction model using historical data from source projects and applies the model to target projects. Most previous efforts assumed the cross-project data have the same metrics set, which means the metrics used and the size of metrics set are the same. However, this assumption may not hold in practical scenarios. In addition, software defect datasets have the class-imbalance problem which increases the difficulty for the learner to predict defects. In this paper, we advance canonical correlation analysis by deriving a joint feature space for associating cross-project data. We also propose a novel support vector machine algorithm which incorporates the correlation transfer information into classifier design for cross-project prediction. Moreover, we take different misclassification costs into consideration to make the classification inclining to classify a module as a defective one, alleviating the impact of imbalanced data. The experimental results show that our method is more effective compared to state-of-the-art methods.