Edson OliveiraJr

Despite the advances promoted by the Web Content Accessibility Guidelines (WCAG), images on the Web still frequently lack adequate alternative text, limiting access for people with visual impairments. This paper investigates the use of Multimodal Large Language Models (MLLM) for the automatic generation of image descriptions in journalistic contexts, using domain-specialist alt text authored by journalists as the reference baseline. We examine the impact of contextual information extracted from the page’s HTML and the effect of different word-length constraints on the generated descriptions. A case study was conducted with 73 images collected from the Brazilian news portal O Globo , for which descriptions were generated by two MLLM (Gemini and GPT) under multiple configurations. The generated descriptions were evaluated using BERTScore, CLIPScore, and RefCLIPScore, which measure alignment with both domain-specific alt text and the images’ visual content. The results show that imposing word limits increases semantic overlap with reference texts, whereas including contextual information has a limited, model-dependent effect. Overall, the descriptions produced by MLLM achieved similarity levels comparable to domain-specialist alternatives under textual and hybrid metrics, suggesting that MLLM constitute a promising baseline for large-scale deployment of alternative texts.

The exponential growth of heterogeneous textual data within modern sociotechnical ecosystems—such as healthcare, education, technology, and the social sciences—has surpassed the capacity of traditional qualitative analysis methods to remain agile and precise when synthesizing evidence, analyzing software requirements, or mining issue repositories. Large Language Models (LLMs), driven by recent advances in Artificial Intelligence (AI), have emerged as transformative tools capable of automating and enhancing qualitative analysis. This study extends a previously published systematic mapping of the literature that maps and evaluates the use of LLMs in qualitative research. We extend the prior study by increasing the number of included studies by means of a snowballing process, exploring their application contexts, configurations, methodologies, and evaluation metrics. The original study analyzed eight papers, from which six were used as seeds in the snowballing process, leading to the identification and analysis of 38 additional studies, resulting in a total of 44 included papers. The results reveal that, beyond being applied across diverse fields, LLMs demonstrate the ability to reduce the effort traditionally required in tasks such as coding, categorization, and theme extraction, with 56% of studies reporting performance as either strictly equal (45%) or superior (11%) to manual analysis. However, challenges such as reliance on prompt engineering, occasional inaccuracies, and contextual limitations remain as barriers to their broader use. This research highlights opportunities to integrate LLMs with human expertise to improve model robustness and refine evaluation methodologies.

In July 2024, a failed update in the CrowdStrike Falcon security software disrupted critical services across the United States and several other countries, affecting sectors such as aviation, healthcare, and finance. Despite advancements in automated testing, an error introduced during development was rapidly propagated, exposing vulnerabilities in digital infrastructure and underscoring the risks of technological interdependence. This study examines the incident through an integrative, multivocal literature review that combines academic and gray literature to address three core research questions: the impacts of the outage (RQ1), the technical causes of its severity (RQ2), and the lessons learned for software engineering and cybersecurity (RQ3). The analysis reveals critical failures in validation processes, integration mechanisms, and organizational prac-tices, highlighting the need for robust quality assurance and software governance. The findings contribute to both research and practice by informing more effective strategies for preventing and responding to large-scale cybersecurity incidents.