While the benefits of subtitles for comprehension are widely acknowledged, empirical research has largely neglected the insights offered by user reviews of mobile applications. This study presents a comprehensive analysis of user feedback on subtitle features in the top 230 mobile applications across various categories. We curated a dataset of 48,872 user reviews specifically related to subtitles, manually categorizing them based on user-provided information. To understand sentiment and accessibility concerns, we conducted sentiment analysis and extracted accessibility-related reviews. Our findings offer valuable insights into user challenges and expectations regarding subtitles in mobile applications, providing a foundation for enhancing their design and implementation. These insights include the need for improved customization options, better accessibility features, and more responsive subtitle performance.

Usability is a crucial aspect of software development, yet it is not emphasized enough in computing education. Studies show that poor usability leads to significant financial losses, with businesses in the United States alone incurring 62 billion in annual losses due to subpar user experiences. Despite its importance, usability principles are not sufficiently taught to computing students. This experience report presents HeuristicBuilder, an interactive multimodal educational tool designed to teach Jakob Nielsen's 10 usability heuristics. These principles are essential for designing user-friendly interactive systems, but often pose challenges in practical application. HeuristicBuilder addresses this by combining theoretical learning via reading and quizzes with practical UI design exercises. The tool was evaluated with 224 participants, showing significant improvements in both theoretical understanding and practical application of usability principles. Participants found HeuristicBuilder to be valuable and effective, preferring its interactive approach over traditional methods such as reading books or online courses. In conclusion, HeuristicBuilder proves effective at educating users about usability heuristics through engaging, accessible means. Its positive reception highlights its potential to enhance usability education and user experience design practices. Future enhancements aim to further optimize interactive learning and practical application, advancing skills among interaction designers and usability practitioners.

This experience report presents the development and evaluation of a virtual reality (VR)-based, gamified platform designed to teach accessibility concepts in an engaging and immersive manner. The system integrates interactive scenarios and game elements to enhance student motivation and support knowledge retention. To assess its effectiveness, we conducted a study with 35 students who used the VR platform and provided feedback through a post-intervention survey. The findings indicate positive perceptions of the learning experience, with participants reporting increased awareness and understanding of accessibility principles. Drawing on the design and evaluation process, this report highlights the value of an embodied learning approach that integrates VR and gamification to enhance the impact of accessibility education. It offers practical insights into the design, implementation, and pedagogical potential of VR-based, gamified tools for enhancing accessibility in computer learning education.

As artificial intelligence (AI) technologies become more deeply integrated into everyday life, the ethical challenges they pose—from algorithmic bias and privacy breaches to questions of accountability and societal impact—have gained attention. While there is existing research on how general or tech ethics is taught within computer science (CS) education, comparatively little is known about the specific treatment of AI ethics within CS education. This study addresses that gap through a large-scale analysis of 955 AI-related publicly available course syllabi from top U.S. universities, focusing on the presence and depth of instruction on AI ethics. Of these, only 83 courses incorporated AI ethics content. We analyzed these syllabi to assess the extent of coverage, instructional approaches, topics addressed, learning objectives, and assessment methods used in teaching AI ethics. Our findings indicate that AI ethics is most often integrated into broader AI-related technical courses rather than taught as a dedicated subject, with most courses allotting only one or two sessions for these topics. Commonly addressed themes include algorithmic bias, data privacy, transparency, and social impact. This work highlights both the progress and the gaps in preparing future technologists to engage with the ethical dimensions of AI, and it offers a framework for enhancing the integration of AI ethics in computing curricula.

While numerous studies have explored the integration of accessibility into computing courses, including software engineering courses, most focus on awareness-building or post-development WCAG compliance testing, often overlooking the shift-left principle, which advocates for embedding accessibility considerations early in the software development lifecycle, particularly during the design and planning stages. To address this gap, we redesigned a software engineering course to incorporate accessibility from the outset, beginning with requirements elicitation and UX design. Students received targeted training in using accessibility evaluation tools such as Stark and Axe for Figma during the design phase, followed by continued testing in the post-development (testing) phase using tools like AChecker and WAVE. We analyzed student projects to understand common accessibility issues detected and fixed across phases, and compared issue types across tools. Our findings affirm the value of a multi-tool, multi-phase approach: early design checks enable the identification and remediation of low-effort, high-impact issues, thereby reducing rework and freeing bandwidth for more complex development-level fixes. This work provides practical insights for computing educators aiming to meaningfully integrate accessibility into core software engineering curricula and contributes to a broader cultural shift toward building accessible technologies by design.

Skin-toned emojis are crucial for fostering personal identity and social inclusion in online communication. As AI models, particularly Large Language Models (LLMs), increasingly mediate interactions on web platforms, the risk that these systems perpetuate societal biases through their representation of such symbols is a significant concern. This paper presents the first large-scale comparative study of bias in skin-toned emoji representations across two distinct model classes. We systematically evaluate dedicated emoji embedding models (emoji2vec, emoji-sw2v) against four modern LLMs (Llama, Gemma, Qwen, and Mistral). Our analysis first reveals a critical performance gap: while LLMs demonstrate robust support for skin tone modifiers, widely-used specialized emoji models exhibit severe deficiencies. More importantly, a multi-faceted investigation into semantic consistency, representational similarity, sentiment polarity, and core biases uncovers systemic disparities. We find evidence of skewed sentiment and inconsistent meanings associated with emojis across different skin tones, highlighting latent biases within these foundational models. Our findings underscore the urgent need for developers and platforms to audit and mitigate these representational harms, ensuring that AI's role on the web promotes genuine equity rather than reinforcing societal biases.

Digital accessibility ensures that digital products and services are usable by a diverse range of users, regardless of their physical or cognitive abilities. While numerous standards and guidelines have been established to aid developers in creating accessible content, studies reveal a persistent lack of accessibility in many web and mobile applications. This gap is often attributed to barriers such as lack of awareness, insufficient knowledge, absence of specific requirements, time constraints, and lack of executive support. In this context, we aim to address the lack of awareness and knowledge challenges by proposing a hands-on approach that leverages the capabilities of Large Language Models (LLMs) like ChatGPT to enhance students' accessibility awareness, knowledge, and practical skills. We engaged software engineering students in tasks involving website development and accessibility evaluation using checker tools, and we utilized ChatGPT 3.5 to fix identified accessibility issues. Our findings suggest that practical assignments significantly enhance learning outcomes, as interactions with LLMs allow students to develop a deeper understanding of accessibility concepts. This approach not only reinforces theoretical knowledge but also highlights the real-world impact of their work. The results indicate that combining practical assignments with AI-driven support effectively improves students' proficiency in web accessibility.

This experience report contributes to the expanding body of literature advocating for the inclusion of accessibility topics within core computing courses rather than restricting them to specialized electives such as human-computer interaction (HCI). We integrated an accessibility-focused sprint into a 3-credit software engineering course at a large private university in the US. Throughout the course, students progressively developed a software project in groups with a dedicated sprint designed to emphasize the importance of accessibility. This sprint included lectures on accessibility principles and hands-on accessibility testing on their own projects, followed by a sprint to fix the issues found. A survey was conducted at the end of the course to gauge their learning outcomes and perceptions of accessibility. The results demonstrate that integrating accessibility into a software engineering course not only enhances students' technical skills but also boosts their commitment to creating inclusive software. We present the accessibility interventions made, assessment instruments used and our findings in this report. We also offer key insights as takeaways for the computing education community, providing guidance for educators aiming to embed accessibility into their curriculum.

With over 95% of the top one million websites not being fully accessible, teaching digital accessibility to computing students is crucial. In this study, conducted over three spring semesters within a Software Engineering project course, we introduced dedicated sprints focused on accessibility. During these sprints, students were taught about various types of disabilities, accessibility principles, web content accessibility guidelines (WCAG), and various automated and manual techniques to test for accessibility. Students utilized automated tools to identify accessibility issues in their web projects and subsequently dedicated another sprint to address and resolve these issues. We systematically documented common accessibility mistakes students make, highlighting the WCAG 'success criteria' that require careful instructional focus. Additionally, we identify which accessibility issues are frequently and easily resolved by students and which challenges persist despite their efforts. This comprehensive analysis provides valuable insights, enabling the computing education community to effectively integrate and emphasize accessibility instruction in their curricula, ultimately fostering more inclusive and accessible web development practices.

Accessibility is an important quality factor of mobile applications. Many studies have shown that, despite the availability of many resources to guide the development of accessible software, most apps and web applications contain many accessibility issues. Some researchers surveyed professionals and organizations to understand the lack of accessibility during software development, but few studies have investigated how developers and organizations respond to accessibility bug reports. Therefore, this paper analyzes accessibility bug reports posted in the Chromium repository to understand how developers and organizations handle them. More specifically, we want to determine the frequency of accessibility bug reports over time, the time-to-fix compared to traditional bug reports (e.g., functional bugs), and the types of accessibility barriers reported. Results show that the frequency of accessibility reports has increased over the years, and accessibility bugs take longer to be fixed, as they tend to be given low priority.

Refactoring is a widely adopted practice that keeps code healthy and provides well known benefits like improving developer productivity. Developers routinely make decisions about how to refactor code (which specific refactoring changes to make), but the criteria that guide these decisions is not well studied. We conducted a multi-method study to understand the diversity of criteria that developers use in deciding what refactoring changes to make, the relative importance of different criteria, and the extent to which refactoring recommendation tools incorporate these criteria in their recommendation approaches. Our findings demonstrate that developers in industry situationally employ more than a dozen criteria when making refactoring decisions. However, no recommendation tool supports even half of those criteria and most criteria are supported by only a few tools. While research in refactoring recommendations tools is ripe, lack of support for criteria developers care about leaves industry without the kind of recommendation tools that they need. In this paper, we summarize findings from industry interviews, an industry survey, and an analysis of refactoring recommendation tools. We highlight gaps in refactoring recommendation tools that researchers and tool vendors should consider focusing on for successful practical application of refactoring recommendation tools at scale.

Educating students about software testing practices is integral to the curricula of many computer science-related courses and typically involves students writing unit tests. Similar to production/source code, students might inadvertently deviate from established unit testing best practices, and introduce problematic code, referred to as test smells, into their test suites. Given the extensive catalog of test smells, it becomes challenging for students to identify test smells in their code, especially for those who lack experience with testing practices. In this experience report, we aim to increase students' awareness of bad unit testing practices, and detail the outcomes of having 184 students from three higher educational institutes utilize an IDE plugin to automatically detect test smells in their code. Our findings show that while students report on the plugin's usefulness in learning about and detecting test smells, they also identify specific test smells that they consider harmless. We anticipate that our findings will support academia in refining course curricula on unit testing and enabling educators to support students with code review strategies of test code.

Multi-environment networks, such as those in smart homes, handle both IoT and traditional IP-based traffic. Weak security protocols in IoT devices and the diverse traffic flow make these networks vulnerable to security breaches. This study delves into this pressing challenge and presents a pioneering solution—FAMTDS (Fully Automated Malicious Traffic Detection System)—designed explicitly for multi-environment networks. FAMTDS addresses the critical need for robust security measures by intelligently analyzing the amalgamation of IoT and IP-based traffic. FAMTDS comprises three pivotal stages: innovative multi-environment dataset creation, optimization of machine learning model hyperparameters, and holistic system optimization. For the multi-environment dataset, we amalgamate two prominent open-source datasets, UNSW-NB-15 and IoTID-20. Initially, crucial features are extracted from both datasets using an extra trees classifier. Subsequently, an equal number of features is generated by employing a neural network to harmonize these datasets. The resulting multi-environment dataset encompasses 19 distinct attack types, a comprehensive inclusion unprecedented in prior research on malicious traffic detection. This dataset exhibits diversity owing to its varied traffic samples, addressing a crucial gap in existing studies. To accommodate this diversity, machine learning models are deployed with fine-tuned hyperparameters. The Mouth Flame Optimizer streamlines feature extraction, feature generation, and hyperparameter tuning, automating the optimization process. FAMTDS demonstrates exceptional performance, achieving an accuracy score of 0.85 for the multi-environment dataset. We also integrated the CICDDOS2019 dataset with IoTID-20 in our multi-environment dataset, achieving a notable accuracy of 0.82 against recent attacks, thus enhancing our approach’s validation. To further validate the generalizability of our proposed approach, we applied it to zero-day attack prediction. Our method demonstrated an accuracy of 0.84 for zero-day attacks, indicating its effectiveness in detecting newly emerging threats in multi-environment networks.

With the large use of social media, the dissemination of intentionally altered and falsified information has become easy, thus posing negative effects on society. Detecting fake content is a non-trivial task as fake news has unique characteristics and challenges. Additionally, the wide use of artificial intelligence (AI) for fake content generation makes the detection of fake content further complicated. Fake news presents engineered content, making it difficult for traditional approaches to comprehend. Existing fake news detection approaches face four problems: lack of robustness, adaptability, limited or no use of auxiliary information, and inability to handle diversity. Fake content diversity introduces the models’ complexities and degrades their performance. Similarly, the accuracy of fake news detection approaches remains low for practical systems. This study focuses on detecting fake news by using an AI-based approach to obtain high accuracy and robustness by using the concept of text transformation into images. It transforms the text into a standard image format which enriches the feature space and boosts the performance of machine learning models. Extensive experiments using two different datasets involving binary and multi-class classification reveal that the proposed approach outperforms existing solutions by yielding superior accuracy. The use of AI approaches helps obtain higher accuracy of 99.70% and 92% for fake news detection using ISOT and LIAR datasets, respectively.

Accessibility reviews collected from app stores may contain valuable information for improving apps accessibility. Recent studies have presented insightful information on accessibility reviews, but they were based on small datasets and focused on general accessibility concerns. In this paper, we analyzed accessibility reviews that report issues affecting users with visual disabilities or conditions. Such reviews were identified based on selection criteria applied over 179,519,598 reviews of popular apps on the Google Play Store. Our results show that only 0,003% of user reviews mention visual disabilities or conditions; accessibility reviews are associated with 36 visual disabilities or eye conditions; many users do not give precise feedback and refer to their disability using generic terms; accessibility reviews can be grouped into general topics of concerns related to different types of disabilities; and positive reviews are generally associated with high scores and negative feedback with lower scores.

The emergence of new network architectures, protocols, and tools has made it easier for cybercriminals to launch attacks using AI-based tools, presenting challenges in network security. To protect such systems, a versatile malicious traffic detection system is required that can identify attacks regardless of the type of traffic coming toward the network. In this paper, a system is proposed that can singly analyze multi-environment traffic (IoT and traditional IP-based) to detect malicious activity. The existing techniques for managing Multi-Environment traffic are inefficient due to the absence of AI utilization. To overcome these issues, the proposed approach generates a novel multienvironment traffic dataset by merging existing network datasets containing both traditional IP-based traffic and IoT network traffic. Synthetic Data Augmentation TEchnique (S-DATE) is also proposed to overcome the problem of imbalanced data distribution in the new multi-environment dataset. The results show that the utilization of S-DATE results in faster machine learning model convergence and an improvement in the detection rate of normal and abnormal traffic. The proposed approach achieves an impressive overall detection rate of 0.991 and is statistically significant compared to other state-of-the-art approaches.

ChatGPT OpenAI, a large-language chatbot model, has gained a lot of attention due to its popularity and impressive performance in many natural language processing tasks. ChatGPT produces superior answers to a wide range of real-world human questions and generates human-like text. The new OpenAI ChatGPT technology may have some strengths and weaknesses at this early stage. Users have reported early opinions about the ChatGPT features, and their feedback is essential to recognize and fix its shortcomings and issues. This study uses the ChatGPT tweets Arabic dataset to automatically find user opinions and sentiments about ChatGPT technology. The dataset is preprocessed and labeled using the TextBlob Arabic Python library into positive, negative, and neutral tweets. Despite extensive works for the English language, languages like Arabic are less studied regarding tweet analysis. Existing literature about Arabic tweet sentiment analysis has mainly focused on machine learning and deep learning models. We collected a total of 27,780 unstructured tweets from Twitter using the Tweepy SNscrape Python library using various hash-tags such as # Chat-GPT, #OpenAI, #Chatbot, Chat-GPT3, and so on. To enhance the model’s performance and reduce computational complexity, unstructured tweets are converted into structured and normalized forms. Tweets contain missing values, URL and HTML tags, stop words, punctuation, diacritics, elongations, and numeric values that have no impact on the model performance; hence, these increase the computational cost. So, these steps are removed with the help of Python preprocessing libraries to enhance text quality and consistency. This study adopts Transformer-based models such as RoBERTa, XLNet, and DistilBERT that automatically classify the tweets. Additionally, a hybrid transformer-based model is proposed to obtain better results. The proposed hybrid model is developed by combining the hidden outputs of the RoBERTA and BERT models using a concatenation layer, then adding dense layers with “Relu” activation employed as a hidden layer to create non-linearity and a “softmax” activation function for multiclass classification. They differ from existing state-of-the-art models due to the enhanced capabilities of both models in text classification. Hybrid models combine the different models to make accurate predictions and reduce bias and enhanced the overall results, while state-of-the-art models are incapable of making accurate predictions. Experiments show that the proposed hybrid model achieves 96.02% accuracy, 100% precision on negative tweets, and 99% recall for neutral tweets. The performance of the proposed model is far better than existing state-of-the-art models.

Software containers are becoming the new state of the art in the industry as they are extensively used to deploy systems. Indeed, the use of containers enables better modularity, reusability, and portability compared to other technologies. As the complexity of software systems is dramatically increasing, it is critical to enable optimal usage of the needed resources to execute them such as memory and CPU. Thus, different scheduling strategies are proposed to select the most suitable nodes to execute a set of containers. For instance, the default strategy in the Docker Swarm kit scheduling framework is based on an equal distribution of the containers between nodes independent of their sizes and consumed resources. However, balancing the containers’ workload is a complex problem due to the conflicting objectives of minimizing the number of selected nodes, minimizing the number of containers per node, the number of changes compared to the original schedule, and the coupling between containers allocated to different nodes. To deal with those conflicting scheduling objectives, we propose a scheduler based on a many-objective optimization approach for scheduling the execution of containers between multiple nodes. The proposed approach aims at finding the best allocation for containers in nodes that leads to efficient utilization of resources. To evaluate our approach, we compared the performance of multiple many and multi-objective techniques based on NSGA-II, NSGA-III, and IBEA algorithms using 48 Docker-related systems and the results show that NSGA-III outperforms the other algorithms in quality attributes as well as in CPU, Memory and Network usage.

In recent years, mobile accessibility has become an important trend with the goal of allowing all users the possibility of using any app without many limitations. User reviews include insights that are useful for app evolution. However, with the increase in the amount of received reviews, manually analyzing them is tedious and time-consuming, especially when searching for accessibility reviews. The goal of this paper is to support the automated identification of accessibility in user reviews, to help technology professionals in prioritizing their handling, and thus, creating more inclusive apps. Particularly, we design a model that takes as input accessibility user reviews, learns their keyword-based features, in order to make a binary decision, for a given review, on whether it is about accessibility or not. The model is evaluated using a total of 5,326 mobile app reviews. The findings show that (1) our model can accurately identify accessibility reviews, outperforming two baselines, namely keyword-based detector and a random classifier; (2) our model achieves an accuracy of 85% with relatively small training dataset; however, the accuracy improves as we increase the size of the training dataset.

Test smells are defined as sub-optimal design choices developers make when implementing test cases. Hence, similar to code smells, the research community has produced numerous test smell detection tools to investigate the impact of test smells on the quality and maintenance of test suites. However, little is known about the characteristics, type of smells, target language, and availability of these published tools. In this paper, we provide a detailed catalog of all known, peer-reviewed, test smell detection tools.