8333 research outputs found
Sort by
Exploring Accounting as a Career Choice: A Qualitative Exploration of the Irish Perspective
No full textThere has been a lot of work done in the area of career choice, however, mostly using quantitative methods to draw a connecting line between pre-determined variables and the choice of a career. This study explores the factors influencing the choice of accounting as a career using Ireland as the focal point for the investigation. With the Social Cognitive Career Theory as the theoretical underpinning, this study gathered data following the qualitative method with a semi-structured interview, sampling the opinions and perspectives of ten (10) post-graduate students living and studying in Ireland. The use of the qualitative method helped to provide extra context for the choices of respondents although time limitations impacted the number of respondents.
Using thematic analysis, this study set out to explore the impact of an individual’s perception of their self-efficacy, career outcome expectations and other external determinants on their choice of accounting as a career and discovered three(3) main themes of self-efficacy beliefs, socio-cultural determinants and job related determinants, within which sub-themes of identity, perception of personal strengths, mimetic pressure, familial pressure, career goals, prestige, salary, demand for accountants and opportunities for exposure were found as presented by table 1.0.
This study is positioned to add to literature on the subject of career development and for employers, encourage the development and implementation of measures to attract and retain motivated talent. For educators and career guidance professionals, findings from this study can help in matching students to careers that best fit them
Multimodal Depression Detection using Audio and Visual Features
No full textDepression is one of the most common mental health disorders, yet the diagnosis for this is either not readily available or often misdiagnosed. Recent studies using machine learning techniques for depression detection have shown promising results but more research needs to be done in using multiple modalities of data for depression detection. To address this, this research outlines a BiLSTM with Attention layer model with two separate pathways, one for audio modality and the other for video modality. Further, a comparison is done with single modality models to evaluate the proposed model. The testing was done on the DAIC-WOZ dataset and the proposed model was able to achieve an accuracy of 0.934 and recall of 0.944
Metropolitan City Transportation Analysis & Optimal Route Suggestions
No full textUrban center’s economic prosperity is largely dependent on the development of strong transit systems that enable easy access to key locations such as places of employment, recreation areas, and educational institutions. It will help the environment and reduce the need for personal automobiles. The reliability of transport networks has been studied in the past using stops density. The primary focus of this study is to do a comprehensive analysis of the Dublin transportation system using GIS with spatial analysis and suggest the optimal route for the city. The dataset for this problem is taken from a public source that contains the information on the bus routes. I have formulated the problem into a graph optimization problem by considering bus stops as nodes and the value of the edges as the distance between the nodes(bus stops). I have implemented Dijkstra and A* algorithms to find the optimal route. The Dijkstra algorithm shows a low-cost value of 422.87 compared to the A* algorithm which shows 430.74. On the other hand, Dijkstra visited 152 nodes to find the optimal route while A* visited just 59 nodes. The cost-value difference is quite nominal compared to the node visited which makes the A* algorithm the best performer
Comparative Analysis of Transformer Models for Multi-Class Text Classification
No full textThis paper makes a comparative evaluation of five state-of-the-art transformer models in multi-class emotion recognition: BERT, GPT-3.5, RoBERTa, XLNet, and DistilBERT. Motivated by the demand for detecting emotions with accuracy in so many applications today, this research aimed at comparing these models on accuracy, precision, recall, and F1 score on classifying texts into multiple categories of emotions.
The research employed the usage of the GoEmotions dataset, which is a dataset containing 58,000 Reddit comments with 27 different annotated emotions and consolidated into three major classes, i.e., positive, neutral, and negative. The methodology in this research undertook preprocessing for the dataset, model implementation, and fine-tuning, ending up at the point of developing a comprehensive evaluation framework.
Key findings were that there did exist a performance hierarchy and, quite unexpectedly, DistilBERT outstripped all larger models, scoring 95.88%. Following were RoBERTa, XLNet, BERT, and GPT-3.5, performing in descending order. For all models, in comparison to the neutral or negative ones, recognizing positive emotions was easier. A remarkable exception was GPT-3.5, which, though doing splendidly elsewhere in NLP applications, underperformed in the given task.
This paper aids in adding to this literature by disputing the commonly held belief that improvements in NLP tasks are made when the model's size is increased and focusing on the compression methods of models. The findings have implications for academic research in NLP and for the practical applications of emotion recognition systems, mainly scenarios related to high computational efficiency
Depth Estimation for indoor environments using Augmented and Regularized Data through Knowledge Distillation
No full textDepth Estimation is one of the important applications of computer vision which are further used in autonomous vehicles, robotics vision and AR/VR world. This research focuses on increasing generalization capabilities of depth estimation models on indoor settings which have low lightings, clustered and occluded objects and overall lack the diversity in terms of texture, has consistent and repetitive structural geometry. We deployed teacher-student framework to implement a ResNet-based pre-trained model as the teacher which will generate it’s own pseudo depth maps from NYU-Depth V2 and Augmentations. The student model DenseDepth-169 based on U-Net learns from the teacher model and it’s predictions. The proposal addresses overfitting and generalization problem by employing data augmentation and dropout regularization and increasing overall dataset size significantly. Edge Detection and contrast adjustment further aid in improving input feed quality. The research also provides a base for scalable and efficient indoors depth estimation models that are adaptive to diverse environments
Beyond Accuracy: A Comparative Analysis of Recommendation Models Incorporating Quantitative and Qualitative Evaluation
No full textRecommendation systems plays a vital role in content delivery, user accusation and user retention in various domains. This research develops and compares two recommendation models – embedding-based ranking model and a behavioural pattern learning model – using data from the Findups Daily news application. This study evaluates these models on quantitative metrics and evaluation criteria analysis such as diversity, scalability and cold-start problem. This research underscores the importance of aligning recommendation systems with application-specific needs to optimize user engagement and satisfaction
AI-Driven Test Case Generation and Optimization
No full textThe aim of this research is to develop an AI-Driven model to enhance efficiency and effectiveness of software testing by generating and ordering testcases using Natural language Processing (NLP) and Reinforcement Learning (RL) techniques. The traditional software testing methods are time-consuming and require significant manual effort, which often leads to inefficiency and missing test coverage. This study utilizes NLP to automatically extract test cases from software requirements documents and applies RL to order test execution sequence. The integration of these technologies aims to maximize test coverage, improve testing efficiency and saving time. Through automated test case generation and optimization, this research aims to reduce test execution time and enhance test coverage, thereby supporting more reliable and efficient software development practices. The findings from this study highlight the potential impact of combining NLP and RL in automating software testing process, promising substantial improvements in software quality assurance and development workflows
Enhancing Cloud Storage Security and Efficiency through Integrated Ranked Keyword Search and Cryptographic Techniques: A Multi-Client Approach
No full textCloud storage services usage is growing to handle and store large amounts of data. This demands for advanced security measures that does not compromise data retrieval efficiency. Conventional encryption technologies even though secure, makes it difficult to perform effective searches, that results in trade-off between security and usability. This research work focuses on improving data retrieval relevance while maintaining strong security measures in a multi-client environment by integrating distributed point function (DPF) with frequency-based ranked keyword search. The system is implemented using Python and evaluated using a dataset that contains keywords of 4,000 documents. Key findings from this study shows that the frequency-based ranking algorithm greatly increases search relevancy while having minimal impact on overall efficiency. The system also maintains robust security by having low false positive and false negative rates. This work contributes to the development of more secure and efficient cloud storage solutions, addressing the growing needs of industries like healthcare and finance. Further research can focus on improvising on the scalability of the system and refinement of the ranking algorithms to better handle diverse and large-scale data environments
Improving Load Balancing in Cloud Computing to Minimize Response time and enhancing resource utilization by using Hybrid metaheuristic Ant Colony Optimization-Simulated Annealing Algorithm
No full textIn cloud computing load balancing and task scheduling is critical and is used for the efficient distribution of workloads and computing resources among virtual machines. Load balancing is an NP-hard optimization problem. Unequal distribution of tasks among can lead to underloaded or overloaded VMs leading to poor resource utilization This research explores a hybrid meta-heuristic algorithm for task distribution and load balancing named ACO-SA, which combines the Ant colony optimization for exploring the solution search space and the Simulated annealing algorithm for Exploiting the search space and refining the solution. ACO is a bio-inspired meta-heuristic algorithm and mimics the foraging behavior of ants to find the shortest past to food representing the best way to distribute tasks among VMs. On other hand Simulated annealing algorithm is used to refine the search space for solutions and converge towards a near-optimal solution. The objective function(fitness function) is defined to minimize response time in ACO and SA. Response time and resource utilization is considered as evaluation parameters. The simulation was performed using the Cloudsim toolkit. The simulation results of the proposed ACO-SA algorithm showed improved performance in minimizing response time by 7% compared to traditional meta-heuristic algorithms such as ant colony optimization(ACO) and Particle swarm optimization(PSO). The hybrid algorithms combines the strength of ACO with SA to optimize load balancing in cloud computing
Enhancing Cybersecurity Posture through Web-based Automated Google Dorking
No full textThe growing complexity of cyber threats drove the realization that the tools used in vulnerability assessment had to be more efficient. One of the principal tools, used only by security researchers, was identified as Google Dorking. Executed manually, it was slow and error-prone. The WAGDT - Dorkinator project was undertaken to address the disadvantages of manual methods by introducing a user-friendly automated solution. The tool was tested for effectiveness by comparing it with a different existing tool in terms of accuracy, efficiency, and user satisfaction. The Dorkinator project was developed to provide augmentation in digital reconnaissance by decreasing manual input and increasing security posture. While designing the tool, emphasis was given to automated complex Google Dorking queries. Thus, it would simplify the information-gathering phase in penetration testing. Implemented using efficient and scalable technologies at the time of its development, the tool focuses on user-friendly design and real-time data processing. In terms of evaluation, Dorkinator demonstrated higher usability, faster execution of queries, and high satisfaction compared to the existing tool, Investigator. The results put forward the potential of Dorkinator to change the state of cybersecurity practice by democratizing the process of vulnerability discovery. This work contributed to the automation of more advanced search techniques and is expected to lead to enhancements in the detection of vulnerabilities. It was acknowledged, however, that further testing in other varied environments and the integration of other advanced features, such as machine learning, were yet to be done. Commercial viability, in some sense, is already very plausible, suggesting further avenues of development and application