With the rise of AI in education, knowledge tracing (KT) has become important for modeling students’ knowledge from interaction data. However, existing methods still face three major challenges, including limited modeling of personalized exercise–concept relations, low robustness to noisy interactions, and inefficient training due to suboptimal data selection. To address these issues, we propose R 2 GCurL, a novel KT framework with two key designs. First, we recast KT as a graph classification problem and construct dynamic graphs from student responses, enabling the model to capture structural relations between exercises and concepts for more personalized KT. Second, we introduce a data-centric curriculum learning strategy based on dynamic graph entropy. Under our definition, pairwise dynamic graph entropy measures graph-transition continuity, where larger values indicate stronger structural similarity. Its sequence-level aggregation is used to derive a structure-aware difficulty signal for sample scheduling. On top of this, an RL-based scheduler further adapts batch selection based on model feedback and is especially beneficial under noisier and more unstable training regimes. Theoretical analysis shows that R 2 GCurL has lower computational complexity than existing graph-based KT models. Extensive experiments on five real-world datasets confirm its effectiveness, robustness, and generalizability, including as a plug-and-play enhancement for sequence-based KT models.

ABSTRACT Software development effort estimation (SDEE) is essential for effective project planning and relies heavily on data quality affected by incomplete datasets. Missing data (MD) are a prevalent problem in machine learning, yet many models treat it arbitrarily despite its significance. Inadequate handling of MD may introduce bias into the induced knowledge. It can be challenging to choose optimal imputation approaches for software development projects. This article presents a novel incomplete value imputation model (NIVIM) that uses a variational autoencoder (VAE) for imputation and synthetic data. By combining contextual and resemblance components, our approach creates an SDEE dataset and improves the data quality using contextual imputation. The key feature of the proposed model is its applicability to a wide variety of datasets as a preprocessing unit. Comparative evaluations demonstrate that NIVIM outperforms existing models such as VAE, generative adversarial imputation network (GAIN), ‐nearest neighbor (K‐NN), and multivariate imputation by chained equations (MICE). Our proposed model NIVIM produces statistically substantial improvements on six benchmark datasets, that is, ISBSG, Albrecht, COCOMO81, Desharnais, NASA, and UCP, with an average improvement in RMSE of 11.05% to 17.72% and MAE of 9.62% to 21.96%.

ABSTRACT In the realm of software development, selecting the appropriate Java application programming interfaces (APIs) from a vast pool remains a significant challenge for developers. This research addresses this complexity by tackling the limitations of current API recommendation methods, which often struggle to align API suggestions with the specific queries and development contexts. In this paper, we introduce a novel prompt method named DSKIPP (Development Scenario, key Knowledge and Intention's Progressive Prompt), designed to enhance the efficiency of large language models (LLMs) in Java API recommendations. Firstly, we devise an overview of DSKIPP which conducts LLMs through a sequential process: first, inferring the package level, followed by the class level, and ultimately the method level as an API comprises three distinct components at varying levels—package, class and method. Secondly, at each level, DSKIPP assists LLMs in deducing the development scenario associated with a query and the essential key knowledge relevant to that scenario. This approach enables LLMs to gain a more profound contextual understanding of the query's intention. Moreover, during the inference process at the class and method level, we implement a self‐check mechanism enabling LLMs to validate the results and ensure a more reasoned and reliable outcome. To validate the efficiency of DSKIPP, comparison and ablation experiments are both conducted within Java programming environment. The comparison results affirm that our method outperforms the current state‐of‐the‐art technologies in API recommendation tasks, while the ablation results shed light on why DSKIPP can enhance the reliability of API recommendations in LLMs. This research contributes to the field by offering a more reliable and context‐sensitive solution for API recommendation in software development.

Modeling and predicting the performance of students in collaborative learning paradigms is an important task. Most of the research presented in literature regarding collaborative learning focuses on the discussion forums and social learning networks. There are only a few works that investigate how students interact with each other in team projects and how such interactions affect their academic performance. In order to bridge this gap, we choose a software engineering course as the study subject. The students who participate in a software engineering course are required to team up and complete a software project together. In this work, we construct an interaction graph based on the activities of students grouped in various teams. Based on this student interaction graph, we present an extended graph transformer framework for collaborative learning (CLGT) for evaluating and predicting the performance of students. Moreover, the proposed CLGT contains an interpretation module that explains the prediction results and visualizes the student interaction patterns. The experimental results confirm that the proposed CLGT outperforms the baseline models in terms of performing predictions based on the real-world datasets. Moreover, the proposed CLGT differentiates the students with poor performance in the collaborative learning paradigm and gives teachers early warnings, so that appropriate assistance can be provided.

API, namely application programming interface, can help developers implement their functions conveniently. To implement a function like dialing in Android, developers sometimes need to call many APIs organised in a special pattern, called API call sequence. However, existing methods rarely focus on code generation for API call sequence. In this paper, we introduce neural network into the application field of generating API call sequence code for Android. The purpose is realising Android code automatically generation by inputting function description. To reach this goal, we first design an API call sequence code graph which is named ACSCG to well represent the Android function code structure and then we convert the ACSCG to API call sequence. Besides, we devise an AI model based on Encoder-Decoder neural network to study the corresponding relation feature of function description and API call sequence. When finishing training the model, one can give a function description to it and generate corresponding API call sequence. After all above has been done, an algorithm is implemented to successfully convert the API call sequence into target code. To verify the efficiency of our method, we collect high quality code from Github and Gitee to build a dataset including 1000 items associated with essential functions in Android such as taking photo, file management, android browser and so on. The experiment shows that our model has a better performance in generating API call sequence code than state-of-the-art technologies.

Software developers need to take advantage of a variety of APIs (application programming interface) in their programs to implement specific functions. The problem of API misuses often arises when developers have incorrect understandings about the new APIs without carefully reading API documents. In order to avoid software defects caused by API misuse, researchers have explored multiple methods, including using AI(artificial intelligence) technology.As a kind of neural network in AI, Transformer has a good sequence processing ability, and the self attention mechanism used by Transformer can better catch the relation in a sequence or between different sequences. Besides it has a good model interpretability. From the perspective of combining API misuse detection with AI, this paper implements a standard Transformer model and a target-combination Transformer model to the learning of API usage information in a named API call sequence extracted from API usage program code. Then we present in the paper the way that our models use API usage information to detect if an API is misused in code. We use F1, precision and recall to evaluate the detection ability and show the advantages of our models in these three indexes. Besides, our models based on Transformer both have a better convergence. Finally, this paper explains why the models based on Transformer has a better performance by showing attention weight among different elements in code.

Software engineering and development is well-known to suffer from unplanned overtime, which causes stress and illness in engineers and can lead to poor quality software with higher defects. Recently, we introduced a multi-objective decision support approach to help balance project risks and duration against overtime, so that software engineers can better plan overtime. This approach was empirically evaluated on six real world software projects and compared against state-of-the-art evolutionary approaches and currently used overtime strategies. The results showed that our proposal comfortably outperformed all the benchmarks considered. This paper extends our previous work by investigating adaptive multi-objective approaches to meta-heuristic operator selection, thereby extending and (as the results show) improving algorithmic performance. We also extended our empirical study to include two new real world software projects, thereby enhancing the scientific evidence for the technical performance claims made in the paper. Our new results, over all eight projects studied, showed that our adaptive algorithm outperforms the considered state of the art multi-objective approaches in 93 percent of the experiments (with large effect size). The results also confirm that our approach significantly outperforms current overtime planning practices in 100 percent of the experiments (with large effect size).

The nature of the requirements analysis problem, based as it is on uncertain and often inaccurate estimates of costs and effort, makes sensitivity analysis important. Sensitivity analysis allows the decision maker to identify those requirements and budgets that are particularly sensitive to misestimation. However, finding scalable sensitivity analysis techniques is not easy because the underlying optimization problem is NP-hard. This article introduces an approach to sensitivity analysis based on exact optimization. We implemented this approach as a tool, O ATSAC , which allowed us to experimentally evaluate the scalability and applicability of Requirements Sensitivity Analysis (RSA). Our results show that O ATSAC scales sufficiently well for practical applications in Requirements Sensitivity Analysis. We also show how the sensitivity analysis can yield insights into difficult and otherwise obscure interactions between budgets, requirements costs, and estimate inaccuracies using a real-world case study.

Software Engineering and development is well-known to suffer from unplanned overtime, which causes stress and illness in engineers and can lead to poor quality software with higher defects. In this paper, we introduce a multi-objective decision support approach to help balance project risks and duration against overtime, so that software engineers can better plan overtime. We evaluate our approach on 6 real world software projects, drawn from 3 organisations using 3 standard evaluation measures and 3 different approaches to risk assessment. Our results show that our approach was significantly better (p <; 0.05) than standard multi-objective search in 76% of experiments (with high Cohen effect size in 85% of these) and was significantly better than currently used overtime planning strategies in 100% of experiments (with high effect size in all). We also show how our approach provides actionable overtime planning results and investigate the impact of the three different forms of risk assessment.

Software engineering is plagued by problems associated with unreliable cost estimates. This paper introduces an approach to sensitivity analysis for requirements engineering. It uses Search-Based Software Engineering to aid the decision maker to explore sensitivity of the cost estimates of requirements for the Next Release Problem (NRP). The paper presents both single- and multi-objective formulation of NRP with empirical sensitivity analysis on synthetic and real-world data. The results show strong correlation between the level of inaccuracy and the impact on the selection of requirements, as well as between the cost of requirements and the impact, which is as intuitively expected. However, there also exist a few sensitive exceptions to these trends; the paper uses a heat-map style visualisation to reveal these exceptions which require careful consideration. The paper also shows that such unusually sensitivity patterns occur in real-world data and how the proposed approach clearly identifies them.

Requirements engineering for multiple customers, each of whom have competing and often conflicting priorities, raises issues of negotiation, mediation and conflict resolution. This paper uses a multi-objective optimisation approach to support investigation of the trade-offs in various notions of fairness between multiple customers. Results are presented to validate the approach using two real-world data sets and also using data sets created specifically to stress test the approach. Simple graphical techniques are used to visualize the solution space.