Voice assistants have become important mediators of digital information access, yet their evaluation continues to focus primarily on technical performance metrics such as response accuracy and latency. This study examines the retrieval efficiency of three leading artificial intelligence–driven voice assistants; Amazon Alexa, Google Assistant and Apple Siri, using an integrated analytical perspective based on information foraging theory and the technology acceptance model. A total of 30 validated factual queries across six knowledge domains were used to evaluate system performance in terms of response latency, factual accuracy, answer modality and dependence on external sources. The results indicate clear differences across systems. Amazon Alexa demonstrates the highest overall retrieval efficiency, characterized by accurate and direct responses with relatively low latency. Google Assistant shows balanced performance with moderate latency and more elaborated responses, while Apple Siri relies more frequently on external links, introducing additional interaction steps. Interpreted through the integrated information foraging theory–technology acceptance model framework, these findings suggest that system-level performance differences may be understood in terms of interaction efficiency and their potential relationship to user-oriented evaluation. However, such interpretations remain conceptual, as user perceptions are not directly measured in this study. While the analysis is limited to factual queries and controlled conditions, the study contributes by introducing retrieval efficiency as a composite measure for evaluating voice assistant performance and by demonstrating the value of integrating cognitive and behavioural perspectives. The findings offer implications for the design of more efficient and interaction-oriented voice-based information systems.
This exploratory study examines artificial intelligence anxiety among academic librarians in the United Arab Emirates, a multilingual context undergoing rapid digital transformation. A cross-sectional online survey of 72 academic library professionals, recruited through purposive and snowball sampling, employed a newly developed 35-item instrument covering seven conceptually derived dimensions: job role evolution, ethical and bureaucratic concerns, patron resistance and social trust, governance and strategic alignment, infrastructure and digital divide, vendor and sustainability, and linguistic and cultural anxiety. Internal consistency ranged from acceptable to strong (α = 0.774–0.898). Overall, AI anxiety was moderate (M = 3.33 on a five-point scale), with ethical and bureaucratic concerns (M = 3.47), infrastructure and digital divide (M = 3.42), and vendor and sustainability concerns (M = 3.36) recording the highest means. However, the narrow range across construct means (0.23 points) warrants treating these differences as descriptive rather than substantive rankings. Concerns about storing patron data on foreign servers and shortages of skilled staff received the highest item-level ratings. Although women reported higher scores across all seven dimensions, no gender or other demographic differences remained statistically significant after correction for multiple comparisons. The findings highlight the institutional, infrastructural, ethical and governance conditions associated with AI anxiety in academic libraries. The proposed dimensions offer a context-specific framework for further investigation, subject to validation in larger samples. Practical implications include developing localized and multilingual AI tools, expanding inclusive professional development, and strengthening participatory governance of AI procurement and policy.
Scientific innovation increasingly emerges from collaborative research teams, yet the relationship between collaborators’ prior knowledge differences and innovative outcomes remains insufficiently understood. We introduce a large-scale measure of team cognitive distance using transformer-based scientific embeddings derived from more than 37 million OpenAlex publications between 1970 and 2024. Dynamic author-level knowledge representations are constructed from researchers’ prior publication histories, and team cognitive distance is defined as the average semantic distance among coauthors with usable pre-publication knowledge embeddings in the embedding space. We examine how this measure is associated with two complementary outcomes: scientific disruptiveness and combinatorial novelty. Greater team cognitive distance is positively associated with both outcomes, and the relationship remains positive across scientific fields, publication cohorts, and team sizes. The pattern is also observed when disruptiveness is measured over a common 5-year citation window, when nonlinear terms are included, and in a biomedical replication based on PubMed records and PubMedBERT embeddings. Interaction analyses show that the strength of the association varies across organizational and compositional contexts.
Large language model–based translation systems are reshaping multilingual information access as interactive, prompt-driven interfaces that restructure the cognitive, social, and institutional conditions of cross-language information seeking. This theory-building review argues this shift constitutes an emergent mode of information behavior warranting a distinct theoretical construct: Prompt-Mediated Multilingual Information Access. Prompt-Mediated Multilingual Information Access is defined as a mode of information access in which a user employs iteratively refined natural-language prompts to elicit artificial intelligence–mediated translation, transformation, or synthesis of information across a language boundary in response to an evolving, goal-directed information need. Drawing on literature spanning library and information science, human–computer interaction, and cross-language information retrieval, this study (1) positions Prompt-Mediated Multilingual Information Access as a distinct construct extending cross-language information retrieval, interactive information retrieval, and conversational search; (2) presents a five-layer socio-cognitive mediation model; (3) develops five falsifiable theoretical propositions linking prompt literacy, trust calibration, language asymmetry, and institutional mediation; (4) operationalizes prompt literacy as a socio-cognitive competence across four dimensions; (5) provides exploratory empirical evidence through an analysis of 1602 real-world interactions (WildChat-1M) and a 16-model large language model benchmark; and (6) identifies implications for library and information science practice and multilingual artificial intelligence governance.
The h -index has gained popularity as one of the most widely used bibliometric measures of academic productivity and impact, influencing decisions on hiring, promotion, funding allocation and university rankings. However, its methodological simplicity obscures substantial limitations. The index favours long academic careers, overlooks disciplinary differences in citation practices, fails to reflect research quality and originality and assigns equal credit to all co-authors regardless of contribution. Database discrepancies and the inability to distinguish positive from negative citations further reduce its reliability, potentially leading to unfair evaluations of early-career and interdisciplinary researchers. This study provides a comprehensive multi-level analysis of the h -index across disciplines and career stages. Rather than relying on large-scale statistical datasets, it synthesises the available literature and uses illustrative cases from Science, Technology, Engineering and Mathematics, social sciences and humanities to demonstrate how database choice, disciplinary citation norms, authorship practices and academic age influence h -index scores. The analysis is complemented by alternative and supporting indicators, including the m -index, g -index and e -index, field-weighted citation impact, altmetrics, narrative curriculum vitae (CVs) and peer assessment. The findings indicate that overreliance on the h -index promotes a longevity-driven and conservative evaluation culture that may disadvantage emerging researchers and undervalue innovative or interdisciplinary work. Accordingly, the proposed framework combines time-normalised and field-adjusted metrics with transparent authorship assessment and qualitative peer review. No single indicator can adequately represent research excellence across disciplines, career stages and diverse forms of scholarly contribution. This approach aims to support fairer, more accountable and future-oriented research evaluation while promoting innovation, equity, academic motivation and responsible assessment practices.
