March 22 ~ 23, 2025, Sydney, Australia
Lana Do and Tehmina Amjad, Khoury College of Computer Sciences, Northeastern University, Silicon Valley Campus, California, USA
Fine-grained sentiment analysis captures subtle emotional tones in text, offering insights beyond positive and negative classifications. It helps users make informed decisions by revealing nuanced opinions and sentiment intensities in textual data. This paper introduces Sentiment-Enhanced Fine-Tuned DeBERTaV3 (FiTSent DeBERTaV3), a classification model designed for both sentence-level and document-level sentiment analysis. Built upon the DeBERTaV3 architecture, our model incorporates tailored fine-tuning strategies to address the unique characteristics of each dataset. On the Stanford Sentiment Treebank (SST-5), fine-tuning addresses shorter, nuanced texts, while for Yelp Reviews, strategies are adapted for longer, narrative-style reviews. Additionally, the use of attention pooling allows the model to prioritize sentiment-critical tokens, enhancing its ability to capture subtle sentiment distinctions. FiTSent DeBERTaV3 achieved competitive performance, outperforming baselines on both tasks. These results highlight the effectiveness of our approach and its versatility in handling datasets with varying lengths and complexities, which have not been jointly evaluated before.
Fine-Grained Sentiment Analysis, DeBERTaV3, Dataset-specific Fine-Tuning, Sentiment-Focused Attention Pooling, Sentence-level analysis & Document-level analysis.
Xinyi Zhou1, Zihao Luo2, 1The Derryfield School, 2108 River Road, Manchester, NH 03104, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
This research presents an AI-driven mental health platform designed to provide personalized emotional support, journaling, and geolocation-based resource navigation [1]. The app integrates an empathetic chatbot, customizable journaling tools, and a resource locator to bridge gaps in mental health accessibility [2]. User surveys with 10 participants revealed high satisfaction with the chatbot’s emotional support and the journaling feature’s guided prompts, while the resource locator faced occasional accuracy issues. Key challenges, including NLP accuracy, geolocation reliability, and advanced customization, were identified as areas for improvement. By leveraging user feedback and iterative updates, this app has the potential to revolutionize accessible mental health support.
Mental Health, AI-Driven Chatbot, Personalized Journaling, Geolocation Resources, User-Centered Design
Lee Seo-jun1, Choi Seo-yeon1, Oh Ji-soo1, and Gyu Tae Bae2, 1High School of Korea, Seoul, 2University of California, Berkeley
Frequent wildfires are becoming an increasing menace to environments, buildings and people especially in places that have large areas of forest cover. In South Korea around 63% of the total land is forested which renders it an area that is continually and severely threatened by wildfires. Usual methodologies of fire detection using satellite images or ground based observation are uncomfortable since they come with drawbacks such as high costs, long response time, and weather interference. In this paper we describe an efficient fire detection system for UNMANNED AERIAL VEHICLE (VTOL) incorporating Convolutional Neural Networks. This makes it possible to greatly enhance detection performance of the models even in complicated conditions that were in the original design. In simulations that approximate the field situations as closely as possible, real time operations of the optimized algorithm yielded 93 percent of the target detection percentage with 20 percent of false positives and a frame latency of 1.2 seconds. Furthermore, the implementation of the model on a Raspberry Pi within a VTOL drone proved the ability of the system for on demand forest fire surveillance and control. This work delineates the promise of drone based systems for fire detection to supplement existing systems for wildland fire prevention.
Fire Detection, VTOL Drones, Convolutional Neural Networks, Real-Time Processing, Wildfire Surveillance.
Aisheng Wang1, Carlos Gonzalez2, 1Crean Lutheran High School, 12500 Sand Canyon Ave, Irvine, CA 92618, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
The YouTube Summarizer is an AI-powered system designed to extract key insights from lengthy YouTube videos using GPT-4o-mini, Retrieval-Augmented Generation (RAG), and FAISS embeddings [1]. The system retrieves transcripts via the YouTube API, processes them using LangChain, and generates concise summaries optimized for gaming strategies, educational content, and technology reviews [2]. Experiments comparing AI-generated summaries with human-written summaries showed that while the system performed well in structured content, it struggled with highly technical topics. Identified challenges include incomplete transcripts, oversimplification, and domain-specific accuracy gaps, which could be addressed through fine-tuning AI models, improving transcript retrieval, and integrating query refinement options [3]. The results demonstrate the efficiency and potential of AI-driven video summarization, paving the way for future enhancements in accuracy, adaptability, and user interactivity.
YouTube Summarization, Artificial Intelligence (AI), GPT-4o and RAG, Video Transcript Processing, FAISS Embeddings.
Abol Froushan, Fellow of the Royal Society of Arts, London, UK
The advent of AI-assisted creativity introduces new possibilities for poetic generation, yet it remains unclear how AI poetry compares to human-authored verse in depth, resonance, and complexity. This paper presents the Poetic Intensity Measurement Framework (PIMF), a structured method for assessing poetic intensity across 15 dimensions, including creative imagination, unpredictability, emotional intensity, and sonic quality. By applying vector space modelling, we compare AI and human poetry, revealing key deficiencies in AI-generated verse, particularly in metaphorical depth, emotional resonance, and unpredictability, which hinder its ability to achieve poetic transcendence. The framework extends beyond poetry, offering applications in music, visual arts, and cinematic expression, demonstrating how poetic intensity manifests across artistic disciplines. Our findings reveal the limits of AI in replicating poetic transcendence while showcasing its potential as a collaborative tool for creative exploration. This research bridges quantitative analysis and humanistic interpretation, contributing to computational poetics, AI creativity, and machine-assisted literary analysis.
Computational Poetics, Poetic Intensity Measurement, AI-Generated Poetry, Creative Imagination in AI, Human-AI Collaboration in Literature.
Peter Luh1, Alan Wilhelm2, 1Retired Physicist, San Jose, California, USA, 2CTO @ Referential.ai, San Francisco, California USA
AI chatbots promise indispensability, yet no standard measures this quality beyond innovation or ethics. Inspired by Samuel Johnsons quip, "Im lost without my Boswell, often attributed via Holmes to Watson, we propose the Boswell Test, a framework assessing AI companions indispensability through mentor-level expertise and intimate user insight. Our initial test, probing complex AI policy queries, reveals strengths in knowledge delivery (such as U.S., China) but gaps in personalization (such as EUs broad ethics, Indias scale focus). We automate such queries via multiple chatbots with cross-assessment of grading each others responses. Automation easily extends to multiple domains. True indispensability where users feel lost without my chatbot, however, requires understanding the human host, an elusive frontier for todays AI, constrained by data and algorithmic limits. .
Boswell Test, Boswell Quotient, Chatbot, Indispensability, Turing Test.
Chenhang Christopher Zhang1, Andrew Park2, 1Concord Academy, 166 Main St, Concord, MA 01742, Computer Science Department, 2California State Polytechnic University, Pomona, CA 91768
With the growing risks for children online, this project introduces a Chrome extension that leverages AI-driven text and image classification to filter harmful content in real time. The system employs BERT for text classification and YOLOv8 for image analysis, dynamically blocking inappropriate material while allowing safe content to pass through. Through experimental evaluation, we identified classification weaknesses, prompting the removal of the monetary and social categories due to persistent misclassification. Iterative dataset refinement and model retraining led to significant performance improvements. Performance testing revealed that GPU acceleration is essential for real-time filtering, as CPU-based deployment resulted in substantial delays. Future work will focus on further dataset refinement, model optimization, and multimodal AI integration to enhance efficiency and accuracy. The results demonstrate the viability of AI-powered real-time filtering, offering a customizable and adaptive approach to online safety. This study lays the groundwork for future advancements in automated content moderation, contributing to a safer digital environment for children. .
Online safety, Child internet use, AI content filtering, Parental controls, Harmful content detection, Web filtering, Nonprofit technology.
T. Charaa1, T. Hamdeni2, and I. Abdeljaoued-Tej2, 1University of Carthage, Tunisia, 2University Tunis-El-Manar, Tunisia
Text-guided image editing on real images, particularly in the context of fashion, presents a highly versatile yet challenging task. This process requires that the editing system take as input only the original image and a textual instruction specifying the desired modifications. The system must autonomously identify the regions of the image to be altered while preserving the other characteristics of the original image. In this paper, we present our approach, which leverages state-of-the-art artificial intelligence techniques, including deep neural networks, large language models (LLMs), and advanced methods for image generation and editing, such as Stable Diffusion and InstructPix2Pix. By integrating these models, our system achieves precise interpretation of textual instructions and ensures consistent application of modifications, while maintaining the visual integrity and authenticity of the original image. This framework offers a robust and scalable solution for text-guided image editing in fashion and other domains.
Artificial Intelligence, Computer Vision, Image Editing, Neural Models, Text-Guided Image Editing, Deep Learning, Large Language Models (LLMs).
Abdeen Mustafa Omer, Energy Research Institute (ERI), Nottingham, United Kingdom
The use of renewable energy sources is a fundamental factor for a possible energy policy in the future. Taking into account the sustainable character of the majority of renewable energy technologies, they are able to preserve resources and to provide security, diversity of energy supply and services, virtually without environmental impact. Sustainability has acquired great importance due to the negative impact of various developments on environment. The rapid growth during the last decade has been accompanied by active construction, which in some instances neglected the impact on the environment and human activities. Policies to promote the rational use of electric energy and to preserve natural non-renewable resources are of paramount importance. Low energy design of urban environment and buildings in densely populated areas requires consideration of wide range of factors, including urban setting, transport planning, energy system design and architectural and engineering details. The focus of the world’s attention on environmental issues in recent years has stimulated response in many countries, which have led to a closer examination of energy conservation strategies for conventional fossil fuels. One way of reducing building energy consumption is to design buildings, which are more economical in their use of energy for heating, lighting, cooling, ventilation and hot water supply. However, exploitation of renewable energy in buildings and agricultural greenhouses can, also, significantly contribute towards reducing dependency on fossil fuels. This will also contribute to the amelioration of environmental conditions by replacing conventional fuels with renewable energies that produce no air pollution or greenhouse gases. This study describes various designs of low energy buildings. It also, outlines the effect of dense urban building nature on energy consumption, and its contribution to climate change. Measures, which would help to save energy in buildings, are also presented.
Renewable technologies, Built environment, Sustainable development, Mitigation measures.
Shangchen Sun1, Han Tun Oo2, 1Andrews College, 15800 Yonge Street, Aurora, ON L4G 3H7 Canada, 2California State Polytechnic University, Pomona, CA 91768
This research evaluates a fencing training application that leverages AI and CV2 for pose estimation and real-time feedback. Two experiments were conducted to address critical performance challenges. The first experiment tested the impact of camera angles on pose estimation accuracy, finding optimal results with a front-facing camera and notable deviations at non-ideal angles [1]. The second experiment assessed the system’s ability to detect head positions when users wore fencing helmets, with accuracy dropping from 100% to 87% due to occlusions. Key findings emphasize the need for user guidance in camera placement and algorithmic improvements for handling occlusions. Proposed solutions include incorporating multi-angle calibration, adaptive algorithms, and potential hardware upgrades like depth-sensing cameras [2]. These improvements aim to enhance accessibility, affordability, and precision, making the application a valuable tool for beginners and advanced fencers alike. Overall, this study highlights the potential of AI-driven solutions in democratizing sports training.
Fencing Training, Pose Estimation, Artificial Intelligence (AI), CV2 Detection, Real-Time Feedback.
Varad Joshi and Kesani Hanirvesh, KL University, India
This paper introduces an AI-driven framework for optimizing Continuous Integration (CI) and Continuous Deployment (CD) pipelines in healthcare software systems. These systems demand high reliability, compliance, and operational efficiency due to their critical nature. Traditional DevOps workflows face challenges such as deployment failures, compliance complexity, and human intervention. Our framework leverages machine learning algorithms for anomaly detection, automated compliance verification, and intelligent monitoring to enhance CI/CD pipelines. Key findings from a case study on a hospital management system reveal a 40% reduction in deployment errors and improved system reliability by 25%. This study demonstrates how AI can bridge critical gaps in healthcare DevOps, enabling safer and more efficient software delivery.
DevOps, Continuous Integration, Continuous Deployment, Healthcare Systems, Artificial Intelligence.
Junade Ali, Engprax Ltd, Edinburgh, Scotland, UK
Case studies have shown that software disasters snowball from technical issues to catastrophes through humans covering up problems rather than addressing them and empirical research has found the psychological safety of software engineers to discuss and address problems to be foundational to improving project success. However, the failure to do so can be attributed to psychological factors like loss aversion. We conduct a large-scale study of the experiences of 600 software engineers in the UK and USA on project success experiences. Empirical evaluation finds that approaches like ensuring clear requirements before the start of development, when loss aversion is at its lowest, correlated to 97% higher project success. The freedom of software engineers to discuss and address problems correlates with 87% higher success rates. The findings support the development of software development methodologies with a greater focus on human factors in preventing failure.
Software development methodologies, Agile software, loss aversion, socio-technical systems.
Calla Huayapa Maxgabriel Alexis1, Huaillapuma Santa Cruz Luis Martin1, Maldonado Mamani Ricardo Anibal2, Zapana Yucra Franklyn1, 1Universidad Nacional de Juliaca, Perú, 2Universidad Andina Néstor Cáceres Velásquez, Puno, Perú
The objective is to determine the impact of three work methods on the assembly times of production lines L1, L2, and L3, and thereby establish a standardized method that ensures the most efficient and uniform performance in the production process. The research methodology is based on an experimental design model where three types of work methods are applied under controlled conditions. Assembly times are evaluated through 10 repetitions per production line. The information obtained is analyzed using analysis of variance (ANOVA), which determines the significant differences between the times of each method applied to the production lines. The results indicate significant differences among the production lines, with a p-value of 0.000, which is less than 0.05. In conclusion, the work method has had an important impact, as production line A presented better strategies and thus improved efficiency.
Work Methods, Production Times, Production Lines, Assembly, Industrial.
Ancia Katjiteo and Enock Limbo Simasiku, Department of Applied Education Sciences, Faculty of Education and Human Sciences, University of Namibia, Namibia
This is a fact that needs to be understood why the subject of entrepreneurship is important in preparations of the young and upcoming generation of learner to face the job market opportunity. Traditional assessment approaches are not effective in identifying various and numerous skills and competencies required for entrepreneurship success. To fill this gap, this systematic review synthesizes the existing research on alternative assessments that have been proposed for use in the context of entrepreneurship education with the aim of informing new assessment methods which could enhance students’ achievement of learning outcomes and readiness for entrepreneurship. In the past, the Namibian assessment practices have been more of norm referenced whereby medium of assessment was mostly paper and pencil tests and written examinations which seem to go against the grain of learners’ diverse learning styles, cultural endowments and multiple intelligences. The advantage of the use of alternative evaluation is that it flexible and comprehensive and allows for showcasing knowledge in a range of ways. This research proceeds through a comprehensive literature review and analysis of peer-reviewed publications to select various multiple assessment approaches. Such methods include, but are not limited to, the following: experiential learning projects, business simulations, case analyses, portfolio presentation and pitching. The review analyses the effect of these other forms of assessments in developing the evaluation of self-critical thinking, creativity problem solving skills and the development of an entrepreneurial spirit. Moreover, it examines the various factors that affect the utilisation as well as the acceptance of the other forms of assessment such as preparation of the teacher, availability of resources, and students. From this analysis, the understanding of how alternative assessment could improve the entrepreneurial education curriculum and future studies.
Alternative assessments; Entrepreneurship education; Assessment methods; Experiential learning; Active learning; Student engagement; Simulation games.
Haleema Azra and Iffath Zeeshan, Department of Education, American College of Education, Indiana, USA
The increasing adoption of big data analytics in higher education presents both opportunities for enhancing student outcomes and significant challenges regarding privacy protection and ethical data governance. This qualitative study investigated how educational institutions can effectively balance the implementation of data analytics while maintaining robust privacy protections and ethical standards. Through semi-structured interviews with 30 participants at Greenfield University, including faculty, administrators, and data privacy officers, the research explored current practices, challenges, and potential solutions in educational data analytics. Using thematic analysis, four major themes emerged: ethical framework challenges, privacy protection measures, student rights and consent, and institutional policy implementation. The findings revealed significant tensions between leveraging data for educational improvement and protecting student privacy, particularly in areas of predictive analytics and early intervention strategies. Participants highlighted the inadequacy of traditional consent mechanisms and the need for more transparent data practices. The study identified critical gaps in existing ethical frameworks and emphasized the importance of developing comprehensive guidelines that balance technological innovation with privacy protection. Key recommendations include implementing more robust data governance frameworks, enhancing transparency in data collection and usage, and developing more effective mechanisms for obtaining meaningful informed consent from students. This research contributes to the growing body of literature on ethical considerations in educational data analytics and provides practical insights for institutions seeking to implement data-driven approaches while maintaining ethical integrity and student privacy.
big data analytics, higher education, student privacy, ethical frameworks, data governance, educational technology.
Majlinda KETA1, Valentina SINAJ2, 1Faculty of Social Sciences, University of Tirana, Albania, 2Faculty of Economy, University of Tirana, Albania
This study aims to examine the impact of gender on digital ethics and the use of technology by academic staff at the University of Tirana (UT), with a focus on differences across faculties. With a sample of 315 lecturers from six faculties of UT, this research uses the questionnaire method to collect data regarding the academic staff’s knowledge of digital ethics, use of technology, trust in digital standards and perception of the impact of digital ethics on institutional culture. Data analysis, descriptive statistics and regression analysis were used to test the hypotheses raised. Data processing was carried out using R software. The results of this study will provide valuable insight into identifying factors that may help or hinder digital transformation and the use of digital ethics in the academic environment. The study can also contribute to the development of strategies to improve awareness and implementation of digital ethics within universities.
Digital ethics, gender, Higher education; academic staff
Noman Abdul Rehman, Research Scholar, Swiss School of Management, Bellinzona, Switzerland
With the evolution of Web 2.0 tools, an array of services and environments for communication—like social networking sites, blogs and wikis—there’s a revolutionary change in the traditional teaching-learning milieu in the hyper-connected educational ecosystem. — such platforms promote peer-to-peer sharing of knowledge, which promotes both collaborative learning and digital engagement. Yet the fast-paced integration of these technologies presents hurdles, including cyber security threats, inconsistent content quality, as well as users differences in digital literacy. The study utilizes a mixed methods approach, involving quantitative data from a structured survey (N = 400) and qualitative insights from semi-structured interviews with 20 education technology professionals. Such as the nature of step-by-step integration of Web 2.0 tools, the advantages, disadvantages, proposed strategies and integrating with educational practices including a proposed wide-reaching conceptual framework to encapsulate the relationships between digital engagement vs cybersecurity measures and sustainable learning outcome. More than 85% of respondents said things like engagement and collaboration have improved, while there are still major deficiencies in areas like cybersecurity preparedness & content management. We also highlight actionable recommendations for educational institutions in how to harness the potential of these digital tools while reducing the risks associated with them.
Artificial Intelligence, Digital Transformation, EdTech, Collaborative learning, Cybersecurity
Yuhan Wu2, Ang Li2, 1Margaret’s Episcopal School, San Juan Capistrano, CA 92675, 2California State Polytechnic University, Pomona, CA 91768
This research explores the development of an AI-powered journaling application designed to improve teen mental health by combining wellness journaling, AI-generated feedback, and trend analysis. The study investigates whether frequent journaling and AI-driven insights contribute to emotional awareness, stress management, and long-term mental well-being [10]. The application uses ChatGPT API to analyze journal entries, providing structured feedback and generating weekly mental health trend reports. Two experiments were conducted: one measuring pre- and post-survey results and another comparing frequent vs. infrequent journaling. Results showed significant improvements in stress reduction and emotional expression among frequent journalers, validating AI-assisted journaling as a promising tool. However, challenges such as AI personalization, user engagement, and external influences require further refinement [11]. This research suggests that AI-powered journaling can serve as a valuable complement to traditional mental health resources, fostering self-awareness and proactive emotional management among teenagers [12].
AI-assisted journaling, Mental health support, Teen well-being, Emotional self-reflection, Hybrid mental health solutions.
Soubhik Baral, Meenakshi Ghai, Kanchan Mali, Nidhi Jain, Central Research Laboratory, Bangalore, India
The demand for smart agriculture is burgeoning worldwide, driven by the increasing need for Internet of Things (IoT) solutions. With India’s status as the world’s most populous nation, there is a significant demand for agricultural advancements. Despite historical advancements in labor efficiency, population growth continues to challenge the balance of supply and demand over time. This has led to a desire for “smart agriculture” which makes use of computer programs for field data analysis and monitoring in addition to sensor-based network deployment. The use of IoT-based smart agriculture technology may benefit family farming and organic agriculture. In this paper, LoRa-based wireless smart network solution was designed and developed. The primary focus of the network solution is the proprietary IoT gateway and sensors, which are interfaced on a Single Board Computers (SBC) to give a wide range of field data collecting capabilities. The study’s key component is its ability to use LoRa-based IoT gateways to gather sensor data from the farm or agricultural field, send it to the localhost LoRaWAN server, and onward via internet to our backhaul network for detailed analysis. To enhance agricultural efficiency, this project developed an Android application capable of monitoring temperature, humidity, wetness, water level, and soil conditions. The network solution will enable farmers and governments to make more informed decisions and optimize the use of resources.
LoRa, IoT gateway, SBC, Sensors.
Joshua Mubemba, Forestry and Environmental Institute of Zambia – FEIZ and Wetlands Forum, Zambia
Environmental compliance and monitoring are crucial for ensuring sustainable development, particularly in resource-intensive sectors such as mining, energy, infrastructure, and agriculture. These sectors often involve large-scale operations that have significant environmental and social impacts, necessitating strict adherence to regulatory requirements to mitigate adverse effects on ecosystems and local communities (Glasson et al., 2019). Effective environmental monitoring plays a pivotal role in ensuring that projects adhere to established guidelines and maintain ecological integrity. However, conventional environmental monitoring methods often suffer from inefficiencies, lack of transparency, and susceptibility to data manipulation (Gupta et al., 2021). Traditional compliance approaches rely heavily on manual reporting, centralized databases, and periodic audits, making them vulnerable to human error, delays, and intentional misreporting (Bolognesi et al., 2020).
Blockchain, Environmental Compliance, Smart Contracts, ESIA, Regulatory Transparency.
Soubhik Baral, Meenakshi Ghai, Kanchan Mali, Nidhi Jain, Central Research Laboratory, Bangalore, India
The demand for smart agriculture is burgeoning worldwide, driven by the increasing need for Internet of Things (IoT) solutions. With India’s status as the world’s most populous nation, there is a significant demand for agricultural advancements. Despite historical advancements in labor efficiency, population growth continues to challenge the balance of supply and demand over time. This has led to a desire for “smart agriculture” which makes use of computer programs for field data analysis and monitoring in addition to sensor-based network deployment. The use of IoT-based smart agriculture technology may benefit family farming and organic agriculture. In this paper, LoRa-based wireless smart network solution was designed and developed. The primary focus of the network solution is the proprietary IoT gateway and sensors, which are interfaced on a Single Board Computers (SBC) to give a wide range of field data collecting capabilities. The study’s key component is its ability to use LoRa-based IoT gateways to gather sensor data from the farm or agricultural field, send it to the localhost LoRaWAN server, and onward via internet to our backhaul network for detailed analysis. To enhance agricultural efficiency, this project developed an Android application capable of monitoring temperature, humidity, wetness, water level, and soil conditions. The network solution will enable farmers and governments to make more informed decisions and optimize the use of resources.
LoRa, IoT gateway, SBC, Sensors.
Keyi Yu1, Jonathan Sahagun2, 1Arcadia High School, 180 Campus Dr, Arcadia, CA 91006, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
Environmental pollution, particularly waste accumulation, is a global challenge requiring innovative solutions. This paper introduces an integrated drone and mobile app system to detect, track, and clean up trash. The program comprises three components: a user-friendly mobile app, AI-powered drones for trash detection and collection, and a centralized database for data storage and tracking [1]. Experiments revealed the systems strengths in waste detection accuracy and operational efficiency, alongside limitations in handling extreme environments and dataset diversity. By building on existing methodologies, the project offers a scalable, adaptable, and proactive approach to waste management. This transformative system has the potential to alleviate environmental degradation, engage communities, and preserve natural ecosystems, making it a compelling solution to a pressing issue.
Trash detection, Volunteer efforts, Drones, Apps.