80 research outputs found
Translation of User Crochet Patterns to CrochetPARADE syntax using Large Language Models
No full textCrochet, with its rich history and popularity, provides a creative and therapeutic outlet for millions across the globe, from many walks of life. However, crochet pattern creation and modification can be challenging for novice users, due to the spatial reasoning and structural understanding of stitches required. CrochetPARADE is a tool created to ease this process through pattern visualisation, but it uses a syntax that differs from standard notation and may not be intuitive to the average crocheter. This study explores the use of Large Language Models (LLMs) to translate user-generated crochet patterns into the CrochetPARADE syntax. The first structured, open-source collection of crochet patterns designed for machine learning applications was created, comprising user-generated patterns and their corresponding CrochetPARADE translations. Various approaches, including baseline, few-shot, and fine-tuning techniques, were evaluated with LLMs. The best results were achieved with fine-tuning DeepSeek-R1-Distill-Llama8b, reaching 74\% accuracy, which has the potential to significantly improve the accessibility and ease of crochet pattern creation for users with varying levels of expertise
Translation of User Crochet Patterns to CrochetPARADE syntax using Large Language Models
No full textCrochet, with its rich history and popularity, provides a creative and therapeutic outlet for millions across the globe, from many walks of life. However, crochet pattern creation and modification can be challenging for novice users, due to the spatial reasoning and structural understanding of stitches required. CrochetPARADE is a tool created to ease this process through pattern visualisation, but it uses a syntax that differs from standard notation and may not be intuitive to the average crocheter. This study explores the use of Large Language Models (LLMs) to translate user-generated crochet patterns into the CrochetPARADE syntax. The first structured, open-source collection of crochet patterns designed for machine learning applications was created, comprising user-generated patterns and their corresponding CrochetPARADE translations. Various approaches, including baseline, few-shot, and fine-tuning techniques, were evaluated with LLMs. The best results were achieved with fine-tuning DeepSeek-R1-Distill-Llama8b, reaching 74\% accuracy, which has the potential to significantly improve the accessibility and ease of crochet pattern creation for users with varying levels of expertise
AI-Driven Usability Testing: Integrating Eye-Tracking Data and Agentic Systems for Automated UI Evaluation
No full textDespite the benefits of user interface/experience (UI/UX) design, traditional usability testing remains resource-intensive and repetitive. This study proposes a novel system that integrates real-time browser-based eye-tracking with a multimodal agentic framework to automate UI evaluation. Participants interacted with task-specific interfaces while their gaze data was captured and analysed by a multi-agent system to generate structured usability reports grounded in heuristic principles. Precision metrics were used to quantify qualitative insights, enabling measurable evaluation. To enhance accessibility, a comparative analysis was conducted between proprietary and open-source Large Language Models (LLMs). Results showed that proprietary models consistently delivered accurate insights, whereas smaller local models struggled with reliability — highlighting future directions for offline deployment. The findings contribute to the advancement of AI-driven solutions in usability evaluation, showcasing how agentic systems integrated with browser-based eye-tracking tools can overcome traditional limitations
Three options for citation tracking: Google Scholar, Scopus and Web of Science
Get PDFBackground:
Researchers turn to citation tracking to find the most influential articles for a particular topic, and to see how often their own published papers are cited. For years researchers looking for this type of information had only one resource to consult: the Web of Science. In 2004, two competitors emerged – Scopus and Google Scholar. The research reported here uses citation analysis in an observational study examining these three databases; comparing citation counts for articles from two disciplines (oncology and condensed matter physics) and two years (1993 and 2003).
Methods:
11 journal titles with varying impact factors were selected from each discipline (oncology and condensed matter physics) using the Journal Citation Reports (JCR). All articles published in the selected titles were retrieved for both years, and a stratified random sample of articles was chosen, resulting in four sets of articles. During the week of November 7-12, 2005 we extracted the citation counts for each research article from the three sources. The actual citing references for a subset of the articles published in 2003 were also gathered from each of the three sources.
Results:
For oncology 1993, Web of Science returned the highest average number of citations, 45.3. Scopus returned the highest average number of citations (8.9) for oncology 2003. Web of Science returned the highest number of citations for condensed matter physics 1993 and 2003 (22.5 and 3.9 respectively). The data showed a significant difference in the mean citation rates between all pairs of resources except between Google Scholar and Scopus for condensed matter physics 2003. For articles published in 2003, Google Scholar returned the largest amount of unique citing material for oncology and Web of Science returned the most for condensed matter physics.
Conclusions:
This study did not identify any one of these three resources as the answer to all citation tracking needs. Scopus showed strength in providing citing literature for 2003 oncology articles, while Web of Science produced more citing material for 2003 and 1993 condensed matter physics, and 1993 oncology articles. Unique material was returned by all three tools. Our data indicate that the question of which tool provides the most complete set of citing literature may depend on the subject and publication year of a given article
Extreme programming implementation in academia for software engineering sustainability
No full textThe practice of making workable and maintainable software to meet the requirements of the use case is best defined as a software engineering practice. Fundamentally this is different from other engineering practices due to the abstraction level involved in the production. This fact produces different approaches for sustainability in the software industry. The importance of this practice towards a sustainable software industry in academia is very important. We propose a framework XPIA (Extreme Programming In Academia) that uses proven industry practices in software engineering with a focus on pair programming which is a practice in extreme programming to build a common knowledge and technical base in academics.</p
Using Generative AI in Participatory Backcasting for Interior and Architectural Design
No full textGenerative AI technologies have been recognised as transformative tools. Their impact on the future of architecture and interior design is seen as inevitable as designers increasingly integrate these tools into various stages of the design process. Research has demonstrated that collaboration between the design team and decision-makers using Generative AI yields better results than a fully automated design process..Backcasting is a planning method that involves envisioning a desired future and then working backwards to plan actions in the present to achieve that future. In a participatory backcasting approach, all stakeholders participate in workshops to develop scenarios and visions that guide present actions, using various tools—such as design, participatory, analytical, and organisational tools—to reach the desired design targets.By employing Generative AI models as one of these planning tools during the participatory backcasting process, stakeholders can collaborate, explore multiple design iterations, save significant time, and allow designers to communicate their ideas to participants from different sectors. As a result, designers channel their efforts on innovation and critical problem-solving.This study used two fine-tuned models based on Stable Difusion and conditional controling neural network called ControlNet to provide additional input and support the designer in achieving their vision. Industry design experts have been asked to evaluate the outputs from these two models and the used input guide of human sketches. This study investigates the integration of AI technologies—such as Stable Diffusion and ControlNet—with the Scribble model in participatory Backcasting in design, aiming to enhance collaborative problem-solving and foster innovative design strategies by combining AI-driven visual synthesis with structured future-oriented planning. Results of this study showed a comprehensive comparison of the two models across three domains was provided, and a framework integrating generative AI into participatory Backcasting workflows was devised to assist designers in various Backcasting approaches.<br/
MASON - A Multi-Agent LLM Framework for No-Code Development
No full textThe rise of No-Code Development (NCD) has enabled citizen developers to build applications without traditional programming expertise. However, as these platforms scale to handle complex, interdependent tasks, their limitations become apparent.Large Language Models (LLMs) offer a potential solution, yet single-agent systems often struggle to manage fullstack development reliably.This study introduces MASON—a Multi-Agent System (MAS) for Open No-code development framework—that coordinates specialized LLM agents via a YAML-based workflow to automate NCD tasks. The system was evaluatedacross four proprietary models—Claude 3.5 Sonnet, GPT- 4o Mini, Gemini 1.5 Flash, and DeepSeek-Chat—using HumanEval and MBPP benchmarks to assess accuracy, executiontime, and stability. MASON configurations showed improvedtask reliability in simpler workflows but introduced latency on more complex tasks. Additional testing with small, locally hosted LLMs revealed significant limitations, emphasizing the need for architectural redesign or model fine-tuningto support deployment in resource-constrained environments
MASON - A Multi-Agent LLM Framework for No-Code Development
No full textThe rise of No-Code Development (NCD) has enabled citizen developers to build applications without traditional programming expertise. However, as these platforms scale to handle complex, interdependent tasks, their limitations become apparent.Large Language Models (LLMs) offer a potential solution, yet single-agent systems often struggle to manage fullstack development reliably.This study introduces MASON—a Multi-Agent System (MAS) for Open No-code development framework—that coordinates specialized LLM agents via a YAML-based workflow to automate NCD tasks. The system was evaluatedacross four proprietary models—Claude 3.5 Sonnet, GPT- 4o Mini, Gemini 1.5 Flash, and DeepSeek-Chat—using HumanEval and MBPP benchmarks to assess accuracy, executiontime, and stability. MASON configurations showed improvedtask reliability in simpler workflows but introduced latency on more complex tasks. Additional testing with small, locally hosted LLMs revealed significant limitations, emphasizing the need for architectural redesign or model fine-tuningto support deployment in resource-constrained environments
EchoScript: Enhancing AI Music Generation for Cinematic Scoring via Script-Aware Fine-Tuning
No full textRecent advancements in artificial intelligence (AI) have significantly transformed the landscape of music generation, enabling context-sensitive and emotionally expressive soundtracks for diverse media applications such as film, gaming, and therapeutic environments. However, existing AI models continue to face persistent challenges in maintaining melodic coherence, thematic continuity, and emotional depth—qualities essential for professional soundtrack production. This research addresses these limitations by fine-tuning MusicGen, a transformer-based generative AI model, to create EchoScript—an optimized variant specifically tailored for cinematic soundtrack composition through script-driven conditioning. A curated dataset enriched with detailed metadata, including genre, mood, instrumentation, tempo, and narrative context, was employed to guide the fine-tuning process. Evaluation results demonstrate substantial improvements over the baseline model. EchoScript achieved a lower Fréchet Audio Distance (FAD) score (4.3738 vs. 4.5492) and outperformed the baseline in structured listening tests, with participants consistently preferring EchoScript for musical quality and narrative alignment. Beyond these empirical findings, the study critically examines technical constraints and outlines key future directions, including symbolic-audio integration, enhanced audio mixing, and the development of standardized evaluation metrics. Collectively, these contributions advance the pursuit of AI-generated music that closely approximates human-level expressiveness and narrative coherence, offering meaningful benefits for creative industries reliant on adaptive and emotionally resonant soundtracks
- …
