Content Categories Research Articles

Can LLMs deeply detect complex malicious queries? A framework for jailbreaking via obfuscating intent

Abstract This paper delves into a possible security flaw in large language models (LLMs), particularly in their capacity to identify malicious intent within intricate or ambiguous inquiries. We have discovered that LLMs might overlook the malicious nature of highly veiled requests, even without alterations to the malevolent text in those queries, thus exposing a significant weakness in their content analysis systems. To be specific, we pinpoint and scrutinize two aspects of this vulnerability: (i) LLMs’ diminished capability to perceive maliciousness when parsing extremely obscured queries, and (ii) LLMs’ inability to discern malicious intent in queries that have been intentionally altered to increase their ambiguity by modifying the malevolent content itself. To illustrate and tackle this problem, we propose a theoretical framework and analytical strategy, and introduce a novel black-box jailbreak attack technique called IntentObfuscator. This technique exploits the identified vulnerability by concealing the genuine intentions behind user prompts, thereby compelling LLMs to inadvertently produce restricted content and circumvent their inherent content safety protocols. We elaborate on two specific applications within this framework: ”Obscure Intention” and ”Create Ambiguity,” which skillfully manipulate the complexity and ambiguity of queries to effectively dodge the detection of malicious intent. We empirically confirm the efficacy of the IntentObfuscator approach across various models, including ChatGPT-3.5, ChatGPT-4, Qwen, and Baichuan, achieving an average jailbreak success rate of 69.21%. Remarkably, our tests on ChatGPT-3.5, boasting 100 million weekly active users, yielded an impressive success rate of 83.65%. Additionally, we verify our approach across a range of sensitive content categories, including graphic violence, racism, sexism, political sensitivity, cybersecurity threats, and criminal techniques, further highlighting the considerable impact of our findings on refining ”Red Team” tactics against LLM content security frameworks.

Comparing the Accuracy of Two Generated Large Language Models in Identifying Health-Related Rumors or Misconceptions and the Applicability in Health Science Popularization: Proof-of-Concept Study.

Health-related rumors and misconceptions are spreading at an alarming rate, fueled by the rapid development of the internet and the exponential growth of social media platforms. This phenomenon has become a pressing global concern, as the dissemination of false information can have severe consequences, including widespread panic, social instability, and even public health crises. The aim of the study is to compare the accuracy of rumor identification and the effectiveness of health science popularization between 2 generated large language models in Chinese (GPT-4 by OpenAI and Enhanced Representation through Knowledge Integration Bot [ERNIE Bot] 4.0 by Baidu). In total, 20 health rumors and misconceptions, along with 10 health truths, were randomly inputted into GPT-4 and ERNIE Bot 4.0. We prompted them to determine whether the statements were rumors or misconceptions and provide explanations for their judgment. Further, we asked them to generate a health science popularization essay. We evaluated the outcomes in terms of accuracy, effectiveness, readability, and applicability. Accuracy was assessed by the rate of correctly identifying health-related rumors, misconceptions, and truths. Effectiveness was determined by the accuracy of the generated explanation, which was assessed collaboratively by 2 research team members with a PhD in nursing. Readability was calculated by the readability formula of Chinese health education materials. Applicability was evaluated by the Chinese Suitability Assessment of Materials. GPT-4 and ERNIE Bot 4.0 correctly identified all health rumors and misconceptions (100% accuracy rate). For truths, the accuracy rate was 70% (7/10) and 100% (10/10), respectively. Both mostly provided widely recognized viewpoints without obvious errors. The average readability score for the health essays was 2.92 (SD 0.85) for GPT-4 and 3.02 (SD 0.84) for ERNIE Bot 4.0 (P=.65). For applicability, except for the content and cultural appropriateness category, significant differences were observed in the total score and scores in other dimensions between them (P<.05). ERNIE Bot 4.0 demonstrated similar accuracy to GPT-4 in identifying Chinese rumors. Both provided widely accepted views, despite some inaccuracies. These insights enhance understanding and correct misunderstandings. For health essays, educators can learn from readable language styles of GLLMs. Finally, ERNIE Bot 4.0 aligns with Chinese expression habits, making it a good choice for a better Chinese reading experience.

Exploring Business Analytics Initiatives in the Accounting Literature: are all accounting areas equal?

Objectives: The goal of our research is to understand why Business Analytics (BA) practices are selected and how they are adopted across various accounting areas. To achieve this, our study maps and analyzes the analytics initiatives documented in the literature that have been adopted by the accounting profession, categorizing these initiatives by accounting area. Method: We conducted a Systematic Literature Review, drawing on publications from the Web of Science and Scopus databases, as well as prominent journals in Information Systems and Accounting Information Systems. Data analysis utilized categorical content analysis with theory- and data-driven codes, aligned with research objectives. Results: Our results provide a comprehensive mapping of Business Analytics literature in accounting, detailing the dimensions - Domain, Orientation, and Technique - by accounting area. This mapping provides a deeper understanding of the relationship between the specific characteristics of each accounting area and the use of Business Analytics. Additionally, we propose a taxonomy based on usage purposes in each area. Finally, we present a research agenda, summarizing key contributions and offering suggestions for future research. Contributons: This research contributes academically by enabling a comparison of Business Analytics adoption and use across different accounting areas, highlighting those with greater maturity in Business Analytics. Additionally, the proposed taxonomy, which considers the purpose of BA within these accounting areas, helps clarify and promote alignment between the accounting discipline and analytics techniques, reducing both conceptual and practical confusion. Furthermore, this research serves as a foundation for accounting professionals to develop their skills in analytics initiatives.

"Does My Kid Have an Ear Infection?" An Analysis of Pediatric Acute Otitis Media Videos on TikTok.

With the rise of social media, online platforms have become a common way to access healthcare information. This study examines the quality of pediatric acute otitis media (AOM) videos on TikTok, a popular short-form video social media platform. A TikTok search was conducted between 8/18 and 8/19/2023 using pediatric AOM hashtags: #pediatric acute otitis media management, #kid ear infections remedy, #child ear infections treatment, and #kid ear infection. Data collected include number of views/shares per day, uploader type (nonmedical influencer, lay individual, and medical professional), and content categories. The Patient Education Materials Assessment Tool for Audiovisual Material (PEMAT-AV) and DISCERN questionnaire measured understandability, actionability, and quality of videos. Multivariable linear regression models were used (significance set at ≤0.05). Of 166 videos, 38.6% (64) of uploaders were medical professionals, and 32.5% (54) were nonmedical influencers. Nonmedical influencer videos were viewed and shared significantly more than those by medical professionals (p < 0.05). Controlling for covariates, physicians were more likely to produce more beneficial and higher quality videos as compared with nonmedical influencers (β = 2.4 and 1.3, p < 0.01, respectively). However, physicians did not have significantly different ratings for understandability compared with nonmedical influencers (β = 0.45, p > 0.05). AOM content on TikTok is often geared toward caretakers of symptomatic children. Although physician-created AOM content was significantly higher quality, these videos reached a statistically smaller audience than those from nonmedical influencers. Addressing misinformation on social media platforms requires physicians to reach larger audiences by producing more actionable and understandable content. NA Laryngoscope, 134:5184-5192, 2024.

Building a culture of content creation: How to maximise your audience engagement

This paper will share how Development and Alumni colleagues at the University of Glasgow built a digital content strategy with the aim of embedding a culture of content creation within Development and Alumni and across the wider university community. The paper will provide easily transferrable insights into how content categorisation decisions can be made and how these decisions can be used to shape an effective and accessible content planner and strategy. The paper will focus on sharing practical tips to build confidence in colleagues and encourage collaboration in all stages of content creation, sharing a case study highlighting the benefits of this approach.

초월영성 상담자 발달 생애사 연구

The purpose of this study is to elucidate the developmental process of transpersonal spiritual counselor. To achieve this, the study applied categorical content analysis, a method of life history research, to five representative transpersonal spiritual counselors in Korea. Four key categories emerged regarding their development as traditional and transpersonal spiritual counselor: 'Experiencing life's suffering and questioning the human mind,' 'Finding a path in psychology and growing as a counselor,' 'Undergoing the developmental process of a transpersonal spiritual counselor,' and 'Achieving a harmonious integration of traditional and transpersonal spiritual counseling.' Those categories included 3, 4, 4, and 3 sub-concepts, respectively. By addressing the relatively underexplored area of transpersonal spiritual counselor development in Korea, this study offers deeper insights into the unique developmental journey of transpersonal spiritual counselors and the lifelong process of counselor growth.

Does YouTube help to reduce plagiarism? A content analysis of YouTube videos and sentiment analysis through viewers comments

Purpose Plagiarism can occur in research in various ways, whether intentionally or unintentionally. This study aims to assess the role of YouTube in providing information and creating awareness about plagiarism through videos. It also investigates the topics related to plagiarism covered in YouTube videos and evaluates viewer satisfaction with the content provided. Design/methodology/approach Using the software Webometric Analyst and various search strings, a total of 526 YouTube videos and 18,050 comments were retrieved on April 23, 2023. The “syuzhet” package has been installed on R-studio to comprehend the sentiment associated with the videos. Findings The greatest number of videos (184) were submitted describing plagiarism software-related information, followed by 122 videos describing examples of plagiarism. The sentiment analysis of viewers’ associated comments revealed that the majority (111.36 %) were positive, indicating that viewers are satisfied with the available videos. Research limitations/implications This study has potential limitations. The number of videos assigned to each content category may be biased to some extent because the videos were not fully viewed and the content categories were assigned manually by analyzing the video title. While analyzing comments for sentiment analysis, some comments are difficult for the software to assign a type of sentiment to, so the percentage of sentiment may be slightly biased. Originality/value All data used in this research are original and verified manually. The data for this study was obtained using software (Webometric Analyst) and cross-checked on a regular basis. The titles of the videos were individually inspected and evaluated to analyze the content and emphasis area of previously posted films. This allows the authors to identify the areas of plagiarism when no or few videos have been submitted.

Is There a Correlation Between Readme Content and Project Meta‐Characteristics?

ABSTRACTContextDevelopers often turn to readme files in GitHub repositories when they intend to contribute, reuse, or extend a project. These files act as a primary source of information, offering insights into various aspects of the repository. The content and organization of readme can have a significant impact on the project's popularity, its development progress, and the growth of its community.ObjectiveWe examine the growth of GitHub repositories through the lens of their meta‐characteristics, which encompass factors like popularity, community engagement, and development progress, to study the relationship of readme files with repository growth. To achieve this, we conduct a correlation analysis to assess the relationship between project meta‐characteristics and the content and organization of a readme file.MethodsTo conduct a correlation analysis between readme files and project meta‐characteristics, we compiled a data set of readme files from 2000 public GitHub repositories, encompassing 10 primary programming languages. We define metrics for popularity, progress, and community engagement meta‐characteristics and analyze the correlation of structural features and categorical content in readme files against these metrics using non‐parametric statistical tests.ResultsThe results are presented from three distinct perspectives of project meta‐characteristics across 10 programming languages. The results reveal a positive and strong association of external reference links, contribution guidelines, and team details in readme files with better popularity and larger community size. However, the influence of contribution guidelines in readme files on rate of progress is not clearly evident.ConclusionThe study was conducted to assess the correlation of readme file content on repository popularity, development progress, and community size. Our experiments revealed positive correlations between readme content and project meta‐characteristics. However, the extent of correlation varied between repositories across programming languages and meta‐characteristics. We discussed the implications of our findings for developers and researchers and have proposed recommendations for repository owners to organize readme files toward increasing the growth of the repositories with better meta‐characteristics.

Factors Favoring the Adoption of Green Finance by Financial Institutions In Dakar: An Exploratory Study

As with any organization, financial institutions, particularly in Dakar, play a crucial role in low-income and emerging economies by mobilizing capital and investing in climate change mitigation. Drawing on neo-institutional theory, this exploratory study aims to understand and identify the factors that may influence the adoption of green finance by banks in Dakar. To achieve this, interviews were conducted with five (5) bank CEOs and three (3) solar energy stakeholders (bank clients), and the data were analyzed using thematic categorical content analysis. The findings, based on the interviewees&amp;rsquo; responses, reveal that top management support, proactive institutional strategies, capacity building, incentive policies, market share development, allocation of funds to green projects, and the digitization of banking services are likely to influence the adoption of green finance by banks in Dakar. Furthermore, the direct involvement of monetary and regulatory authorities, through the development of specific regulations that encourage financial institutions to adopt green finance, is essential.

PAULO FREIRE'S APPROACHES AND PHYSICAL EDUCATION: CHALLENGES AND POSSIBILITIES IN PEDAGOGICAL PRACTICE

The study aimed to analyze the approaches between Paulo Freire and Physical Education, based on the challenges and possibilities in the pedagogical practice of this curricular component. It is a qualitative study, using bibliographic research in periodicals in the field of education and physical education, from 2019 to 2024, using the technique of categorical content analysis by theme for data analysis. It is concluded that there are challenges in bringing Paulo Freire and Physical Education closer together, due to the initial training of teachers, little research production that reflects and treats these themes in their centrality. However, there is the possibility of approaching this knowledge in an organic, systematized and intentional way, with experiences that evidence these approaches based on reflections guided by dialogue, problematization and criticality in basic education and higher education.

Evaluation of the content and quality of schizophrenia on TikTok: a cross-sectional study

Schizophrenia is a common and serious heterogeneous mental disorder that has a significant health and economic impact on families and society. As the prevalence of schizophrenia increases each year, public awareness of the disorder is growing. However, it remains unclear whether the quality of information presented in these videos is satisfactory. Therefore, a comprehensive assessment of the quality and content of schizophrenia-related videos on video-sharing platforms is necessary. We screened 123 schizophrenia-related videos from the video-sharing platform TikTok that met the criteria, and evaluated and analyzed them. First, the basic information provided in the videos was recorded and analyzed. The source and content type of each video was then determined. The educational content and quality of all videos were then evaluated using three rating instruments, Modified DISCERN, Global Quality Scale, and Journal of the American Medical Association (JAMA). Videos from different sources were also compared to see if they were statistically different in terms of educational content and quality. We identified 4 categories of content for the videos. The science introduction category had the largest number of 50, and the least number of videos was etiology and causation at 15. Six categories of sources of videos were also identified: the for-profit organizations category has the lowest number of 10, the health professionals category has the highest number of 45. There was a significant correlation between video source and duration (P = 0.014). The JAMA score was significantly positively correlated with the number of video likes (r = 0.721, P < 0.001). This study evaluated the content and information quality of 123 videos related to schizophrenia on the video-sharing platform TikTok. The findings indicate that while these videos offer some valuable insights into schizophrenia, the overall quality falls short of satisfactory levels, with inadequate reliability and accuracy.

CNN based approach for root zone soil moisture prediction by ground penetrating radar

Abstract Traditional soil moisture prediction methods often face challenges in providing spatially and temporally comprehensive root zone soil moisture (RZSM) with complex soil environment. In this paper, a novel method is proposed for predicting RZSM using images obtained from ground penetrating radar (GPR) data and processed with the convolutional neural network (CNN). The CNN encompasses two major stages: classification and regression. Initially, a CNN is tailored to classify tree root images obtained by GPR into distinct water content categories. Subsequently, through transfer learning, the pre-trained classification network is adapted for regression to predict continuous soil moisture levels. Then, the field experimental results demonstrate the efficacy and robustness of our approach, offering promising avenues for enhancing root zone water management and sustainability.

A Nature-Immersive Virtual Reality Intervention to Support Hospice Family Caregivers: Qualitative Findings from a Pilot Study

ObjectiveFamily caregivers (FCGs) of cancer patients who are in hospice experience psychological challenges and poor quality of life (QOL) as a consequence of caregiving demands and anticipatory loss. Supportive interventions are needed. Exposure to nature offers a source of relaxation; however, FCGs are often homebound and have limited opportunities for respite. Technology-mediated nature experiences delivered via virtual reality (VR) may be an alternative. The purpose was to explore the perspectives of hospice FCGs who completed a nature-immersive VR intervention for the enhancement of their emotional health and QOL. MethodsNine scenes, including a tropical beach, green meadow, underwater ocean, and red savanna, were offered. Participants were asked to select one scene to view for a minimum of 10 minutes for five consecutive days at home. Semi-structured interviews were conducted via videoconferencing software and audio-taped, then transcribed, coded, and analyzed using content analysis. ResultsNine participants (aged 33-76 years old) were interviewed. Two content categories included 1) feasibility and 2) acceptability. Participants reported ease-of-use and denied encountering difficulties. Scenes were relaxing and calming, providing mindfulness and escapism elements. Participants enjoyed the interactive, multisensory experience, offering views of wildlife and nature sounds. Two reported motion sickness and eye strain with longer use. Suggestions included extending the study duration and the option to explore other scenes. ConclusionsFindings provide preliminary evidence for the efficacy of a nature-based VR intervention. Larger, diverse samples, with group randomization, are warranted. Technology-mediated nature experiences offer support for the emotional health and QOL of hospice FCGs. Trial registrationN/A.

‘I ALMOST CRIED’: PRE-SERVICE TEACHER’S EMOTIONAL EXPERIENCES DURING A TEACHING PRACTICUM

Emotions are closely related to teaching and learning activities, especially for pre-service teachers as beginners in the field of teaching. In the classroom, there is not only an exchange of knowledge but also effective aspects of teaching and learning that might be challenging for pre-service teachers as they are mostly unfamiliar with the real classroom situation. Thus, this study aimed to explore and investigate one of the authors’ as a pre-service teacher's emotional experiences during a teaching practicum program, chosen through a purposeful sampling technique. The selected participant met the requirements for the study: an EFL pre-service teacher assigned to a teaching practicum program, capable and willing to write diary entries for data collection. To record the emotional experiences, the participant wrote diary entries during practicum. The data collection will be analyzed using categorical content analysis before writing the findings in narrative form. The findings explicate that pre-service teacher experienced both negative and positive emotions as a result of their interaction related to academic activities and their environment. Both emotions were found to affect pre-service teacher’s strategy, motivation, creativity, goals, students’ learning outcomes, and the whole teaching practicum experience.

Partisanship and Older Americans’ Engagement with Dubious Political News

Abstract Studies based on digital trace data show that older Americans visit and share dubious news sources far more often than younger cohorts, tendencies often attributed to lower levels of digital literacy. At the same time, survey experiments show that older Americans are no worse, if not better, at discerning between false and accurate news. If older Americans can identify misleading news content equally well, why are they still more likely to engage with it in observational settings? In this article, we combine survey measures and digital trace data for three nationally representative samples (N = 9,944) to argue that the existing literature overemphasizes the importance of factors like digital literacy relative to standard political variables such as political interest and partisanship, factors known to increase across the lifespan. Calcified partisanship in particular makes older Americans vulnerable to hyperpartisan news—which is highly slanted but not verified as explicitly false. High rates of engagement with this category of content, which has been examined in survey studies of older citizens less regularly in the literature, may partially explain the high rates of engagement with dubious news domains in behavioral trace data. In all, our findings have important implications for how we understand—and might intervene to reduce—high engagement among this cohort with dubious news.

Viki LibraRy: collaborative hypertext browsing and navigation in virtual reality

ABSTRACT We present Viki LibraRy, a dynamically built library in virtual reality (VR) designed to visualise hypertext systems, with an emphasis on collaborative interaction and spatial immersion. Viki LibraRy goes beyond traditional methods of text distribution by providing a platform where users can share, process, and engage with textual information. It operates at the interface of VR, collaborative learning and spatial data processing to make reading tangible and memorable in a spatially mediated way. The article describes the building blocks of Viki LibraRy, its underlying architecture, and several use cases. It evaluates Viki LibraRy in comparison to a conventional web interface for text retrieval and reading. The article shows that Viki LibraRy provides users with spatial references for structuring their recall, so that they can better remember consulted texts and their meta-information (e.g. in terms of subject areas and content categories).

Predictive Modelling for Sensitive Social Media Contents Using Entropy-FlowSort and Artificial Neural Networks Initialized by Large Language Models

This work offers an integrated methodological framework that integrates the capabilities of large language models (LLMs), rules-based reasoning, multi-criteria sorting, and artificial neural networks (ANN) in developing a predictive model for classifying the intensity of sensitive social media contents. The current literature lacks a holistic consideration of multiple attributes in evaluating social media contents, and the proposed framework intends to bridge such a gap. Three actions constitute the development of the framework. First, LLMs (i.e., GPT4) evaluate the social media contents under a predefined set of attributes, leveraging the power of LLMs in content analytics. Second, rules-based reasoning and multi-criteria sorting (i.e., entropy-FlowSort) determine the categories of social media contents. Lastly, the two previous actions produced a complete dataset that can be used to train a predictive model using ANN to classify sensitive social media contents. With 1100 randomly extracted social media contents and the predefined categories of violations against community standards set by Facebook, the proposed integrated methodology produces an ANN-based classification model with 86.36% prediction accuracy. Comparative analysis using Decision Trees, k-nearest neighbors, Linear Discriminant Analysis, Random Forest, and Naive Bayes classification yields the highest performance of ANN. The predictive model can be used as a decision-support tool to design moderation actions on social media contents.

Writing PISA-Like Mathematics Items: The Case of Tertiary Mathematics Instructors from a State University in the Philippines

&lt;p style="text-align:justify"&gt;Mathematics test items in International Large-Scale Assessments (ILSAs) such as the Programme of International Student Assessment (PISA) and the Trends in International Mathematics and Science Study (TIMSS) are nested in contexts defined in their assessment framework (e.g., the Personal, Occupational, Societal, and Scientific contexts in PISA). This study followed the item-writing activities of four tertiary mathematics instructors in the Philippines as they constructed context-based mathematics items. They were tasked to write four items each, following a set of specifications for PISA content and context categories. The data consisted of transcripts from the focus-group discussion which was conducted days after the task. The transcripts were then analyzed using thematic analysis. The results of this study showed that the phenomenon of item-writing in the context of writing PISA-like mathematics items had two themes: the phases of item-writing and the dimensions of item-writing. Findings showed that the respondents struggled to find realistic contexts and that they engaged in a problem-solving task likened to solving a puzzle as they attempted to satisfy the content, context, and process categories in the table of specifications (TOS). This study contributes to filling in the research gap on item-writing activities, particularly those of mathematics teachers in the Philippines- a country whose recent mathematical performance in the PISA 2018, TIMSS 2019, and PISA 2022 was nothing short of dismal.&lt;/p&gt;

Design of an Educational Psychology Research Material Sharing Platform Based on Fuzzy Clustering Algorithm

In the interdisciplinary field of educational psychology, efficiently managing and accessing research materials remains a formidable challenge due to the vast and varied nature of available data. This paper introduces an innovative platform engineered to streamline the organization and dissemination of educational psychology research materials by harnessing the power of fuzzy clustering algorithms for automated content categorization. Motivated by the imperative to enhance the accessibility and discoverability of scholarly resources, our methodology employs the Fuzzy C-Means (FCM) algorithm, adept at accommodating the intrinsic ambiguity and overlap prevalent in educational psychology topics. The process initiates with a comprehensive preprocessing of diverse research materials, including scholarly articles, reports, and datasets, succeeded by meticulous feature extraction to distill pivotal themes and concepts. Subsequent application of the FCM algorithm facilitates the intuitive grouping of materials into clusters, each with varying degrees of relevance, thereby mirroring the multidimensional essence of the domain. Rigorous experimental validation of our approach underscores its efficacy in augmenting search precision and markedly diminishing the time scholars expend in locating pertinent materials. The platform’s user-centric interface further simplifies the exploration of educational psychology research, offering scholars a streamlined mechanism to access and engage with an extensive repository of resources. Through this study, we demonstrate a significant stride towards overcoming the hurdles in research material categorization and retrieval, heralding a new era of efficiency and accessibility for the educational psychology academic community on a global scale.

Computer Vision Models for Image Analysis in Advertising Research

This study introduces computer vision models for image analysis in advertising research. It reviews the literature in social science and computer science and identifies three categories and nine types of image analysis. The study uses these categories and types as a framework to select 12 computer vision models and compare them on their capability, accuracy, availability, and usability. Nine models are single-functional models, and three are multi-functional models; all 12 have been used in advertising research. The study also demonstrates how two models are used to classify a sample of image ads and assess the aesthetic scores of these ads to answer the research question about the relationship between content categories and aesthetic scores in image ads. It outlines several key steps for the use of computer vision models in advertising research and proposes future research directions. The study can serve as a guide to advertising researchers.