1393 research outputs found
Sort by
Factors Affecting Users’ Satisfaction with Telehealth Apps: Analysis of User Reviews using BERT
Get PDFTelehealth mobile apps and telehealth services are increasingly used by patients, particularly, post-COVID-19.This study examines factors related to users’ satisfaction with these apps and services by analyzing reviews from actual telehealth app users. A total of 53,209 reviews were collected from nine telehealth apps on the Google Play store. Using BERT embeddings, UMAP, and HDBSCAN, topics were generated and labeled to identify these factors. Results showed that telehealth app users expressed several factors related to satisfaction with telehealth apps, which could impact the acceptability and adoption of such apps. These include usability (ease of use), usefulness, convenience and efficiency, cost and affordability, technical performance and connectivity, professionalism and expertise, and comprehensive care support
Real-Time Fake News Detection on the X (Twitter): An Online Machine Learning Approach
No full textFake news, along with the speed of mass communication via social media, is having a significant impact on our social life, particularly in the political world. Fake news detection on social media has recently become an emerging research area that is attracting tremendous attention. Existing studies have proposed to employ traditional machine learning (ML) or emergent deep learning (DL) methods to detect fake news. These ML or DL methods based on batch processing and learning from a training dataset, however, are inefficient in conducting continual real-time fake news detection for incoming unseen social media data. In this research, we propose to use online machine learning (OML) to automatically identify fake news in real time. We investigated various online learning algorithms, including Approximate Large Margin Algorithm (ALMA), Passive-Aggressive (PA), etc., and compared their performance with some existing ML or DL methods for real-time fake news detection on Twitter. The preliminary results of our study demonstrate the considerable potential of OML techniques in classifying real-time fake news, thus highlighting the adaptability and robustness of OML in handling dynamic information streams such as fake news
HLA-G gene polymorphisms as predictors of survival in colorectal cancer: A unified machine learning approach for cancer detection
No full textObjectives
Human Leukocyte Antigen (HLA-G) is a potent molecule involved in immune-tolerance. Here, we investigated the contribution of HLA-G gene polymorphisms (14 bp Ins/Del and +3142C/G) for accurate prediction of colorectal cancer (CRC) overall survival (OS) status. Our study presents a comprehensive investigation of the prognostic value of HLA-G genotypes and haplotypes in predicting OS status in 266 Tunisian patients with CRC. Methods
We used a machine learning (ML)-based framework described below: (1) A dimensionality reduction approach was used to examine evidence of an association between HLA-G genotypes and OS status. (2) Decision-tree ML models were used to explore the performance of the HLA-G genotype as a relevant contributing feature to accurately predict OS status. Results
HLA-G polymorphisms were highly predictive of OS status when a random forest classifier was used. The HLA-G 14 bp Ins/Del polymorphism outperformed the HLA-G + 3142C/G polymorphism as a predictor of OS. The Del/Del genotype was associated with worse OS and the G/G genotype was associated with favorable OS. The InsC haplotype predicted a favorable prognosis, and the DelG haplotype predicted a worse OS. The combined prediction demonstrated, with 100 % precision and high accuracy, that Del/Del genotype associated with key clinical features, can efficiently predict worse OS. The results were evaluated through an external validation process to ensure their reliability
Bridging the Protection Gap: Innovative Approaches to Shield Older Adults from AI-Enhanced Scams
Get PDFArtificial Intelligence (AI) is rapidly gaining popularity as individuals, groups, and organizations discover and apply its expanding capabilities. Generative AI creates or alters various content types including text, image, audio, and video that are realistic and challenging to identify as AI-generated constructs. However, guardrails preventing malicious use of AI are easily bypassed. Numerous indications suggest that scammers are already using AI to enhance already successful scams, improving scam effectiveness, speed and credibility, while reducing detectability of scams that target older adults, who are known to be slow to adopt new technologies. Through hypothetical cases analysis of two leading scams, the tech support scams and the romance scams, this paper explores the future of AI in scams affecting older adults by identifying current vulnerabilities and recommending updated defensive measures focusing the establishment of a reliable support network offering elevated support to increase confidence and ability to defend against AI-enhanced scams
Digital Divide in Rural Native American Homes: A Student Perspective
Get PDFNative American homes located in rural western North Dakota have seen increased availability of broadband and fiber optic Internet but continue to have low access numbers. Native American homes in the region located outside population centers continue to have difficulty accessing the Internet. Many households continue to struggle with economic factors compounded by the cost of a device to access the Internet for personal, educational, or employment use. A qualitative study with semi-structured interviews and the grounded theory method investigated the Native American home environment. Interviews were conducted with Native American students attending a state university with insight into diverse digital environments. The cost of access and technical knowledge continue to be issues in the home. This research deepens the understanding of digital divide factors in Native American households, emphasizing the perspective of Native American students
Natural Language Processing: Understanding Slang and Colloquial Speech
Get PDFWithin the field of Natural Language Processing, slang and colloquial language provide a unique challenge for the training and use of language models. This speech holds an odd, transient space with public consciousness, as each instance has a variable time of relevancy and popularity, amongst other factors such as their fluid definition, usage amongst certain groups, and relative lack of information and training data concerning new or evolving terms. These characteristics make it difficult for models to learn these phrases in a timely manner. This project focuses on how a NLP model could learn new additions to colloquial language in a timely and efficient manner, as well as keep up with a rapidly mutating lexicon.https://scholar.dsu.edu/research-symposium/1035/thumbnail.jp
Leveraging the Zigbee Pattern of Life to Identify Devices
Get PDFZigBee is the open-standard enabling smart devices to be adopted in new and innovative ways. In this research, the network layer of the ZigBee protocol is examined to further the understanding of security impacts it brings to the environment. The first research question is determining of the ZigBee beaconing patterns reveal the device type. This is the first layer that introduces encryption and the results indicate that beacon layer data with the lower layer information do not provide enough information to confidently identify the device prior to admittance to the network. This is important in the identification of rogue devices. The second research question is how applying machine learning to a set of features extracted from network traffic can reveal device types. The results are yes, training a model to identify traffic is possible leveraging network-layer traffic to identify device types within the network. The third research question is revealing if the traffic being captured at the network layer can be categorized as abnormal leading to potential malicious traffic identification. The results indicate a yes when the abnormal traffic is greater than one standard deviation from the average packet size. A reverse majority-vote classifier is created that classifies devices based on observed traffic. This research expands network identification, inventory, and potential detection for ZigBee smart devices that can be leveraged in the field environments today
A Model of Authentication for Augmented and Virtual Realities
Get PDFPasswords have remained the dominant form of knowledge-based authentication (KBA) for greater than five decades. Despite being nearly universally hated by users, practitioners, and researchers, little to no advances have occurred since their first use. This continued use is often attributed to two benefits: speed of completion and cost of implementation. Users are skilled at typing quickly on keyboards, and all major computing systems have been designed to support the keyboard based password. As we move into a new world of computing, specifically in augmented and virtual realities (AR/VR), the value of these benefits is expected to significantly decrease. AR/VR technology has advanced significantly in recent years, with recent applications in military, medicine, and has even the International Space Station. As this technology is poised to enter more mainstream adoption, it is important that security be considered earlier rather than later. A major benefit of AR/VR, is that it provides unrivaled capabilities to create completely new and immersive experiences. Despite these capabilities, the traditional password remains the most observed security artifact, with a virtual QWERTY keyboard being displayed in the AR/VR space. This keyboard, however, is no longer quick and easy to type on with current controller schemes, and the cost of implementing new schemes is drastically reduced from a computational perspective. This research develops a general model of knowledge-based authentication (KBA), and the necessary methods to instantiate this model in an AR/VR context in the form of design principles. These design principles provide methods for successful implementation based on cognitive load theory (CLT). Finally, a fully functional prototype is developed and evaluated assessing both the usability and security properties through a randomized controlled experiment. We find that our abstract model and design principles provide a robust framework for the design and implementation of novel KBA schemes in AR/VR contexts. Additionally, we find that novel mechanisms to provide sequencing and graphical chunking of the users selections enhances memorability and security of user selected knowledge based authenticators This research will draws attention to the importance of considering security artifacts in the context for which they are being used, not for which they were originally developed. This additionally provides guidance towards future advances in knowledge-based authentication in AR/VR, as well as future computing technologies
Dark Side of GenAI: A Blackbox Analysis of X
No full textRecent advancements in generative artificial intelligence (GenAI) have raised many fears, risks, and concerns (Kim 2023; Okey et al. 2023). To shed light on the dark side of GenAI, we collected 55,916 posts from X (formerly Twitter). Based on the content of these posts, we manually labeled a sample set with the corresponding dark side, then identified a short, comprehensive list of GenAI dark sides. Using this list, we trained the ReadMe classifier, a supervised learning algorithm on Brandwatch (“Crimson Hexagon and Brandwatch” 2020), to classify the remaining posts. Further analysis, including emotion analysis and analysis of professions and interests yielded several insights. We found that most posts have a negative sentiment with a total count of 20,777 (89.9%) posts. The emotion analysis showed that majority of the users expressed anger with 15,688 posts (43.4%), followed by fear (7,307 posts, 20.2%), sadness (5,790 posts, 16%), joy (4,255 posts, 11.8%), disgust (2,896 posts, 8%), and surprise (252 posts, 0.7%). Regarding professions, the top four groups expressing concerns about the dark side of GenAI, were the executives group (2,703, 30%), followed by the artists group (1,305, 14%) and software developers and the IT group (1,151 ,13%), then the teachers and lecturers group (1,019, 11%). Similarly, based on interest, the top four groups expressing concerns about the dark side of GenAI, were the technology group (4,446 posts, 18%), followed by those interested in business (3,605 posts, 15%), those who are interested in books (2,831 posts, 11%), and those who are mainly interested in family and parenting (1,892 posts, 8%). Regarding the identified dark sides of GenAI, overall, we identified seven dark sides. Most posts discussed concerns about misinformation and digital deception (18,745 posts, 53%), followed by degradation in quality (8,981 posts, 25%), plagiarism (2,004 posts, 6%), job losses due to automation (1,618 posts, 5%), security and privacy concerns (1,516 posts, 4%), bias (1,454 posts, 4%), ethical concerns (7,92 posts, 4%), and legal and defamation issues (455 posts, 1%). This research not only reported the different types of dark sides discussed on X but also ranked the most discussed topics according to the volume of posts, interests, and professions
Exploring LoRaWAN Class B and Class C Devices: Performance Analysis and Parameter Optimization Strategies
No full textLoRaWAN is an open networking protocol used to create long-range communication for low-powered devices. LoRaWAN allows the configuration of several parameters that affect the network performance. LoRaWAN end devices utilize this protocol for communication in the Internet Of Things (IoT). These devices transmit and receive data through radio gateways using LoRa RF Modulation. LoRaWAN end devices include Class A, B, and C operational categories that vary based on their functionality in communication with the network. Research has been undertaken on Class A end devices, but a notable gap exists in studies dedicated to Class B and Class C devices. Understanding the communication behaviors of these devices is important for finely tuned and efficient implementations. This study aims to bridge this gap by performing a simulation-based performance analysis in ns-3 for these device classes. Additionally, we demonstrate the potential for significant performance improvements through the careful adjustments of system parameters