67471 research outputs found
Sort by
Exploring the Association between Livestreamers’ Self-identified Gender and Their Viewers’ Linguistic Behavior
No full textThis study explores how the gender of streamers on the livestreaming platform Twitch is associated with their viewers’ linguistic behavior. Based on prior literature, we specifically examine how Twitch viewers’ use of two contrasting language types—polite words and swear words—varies by their streamer’s gender. We identified eight gender-diverse Twitch streamers who played a video game together and livestreamed their game play separately on their own channels. From these eight channels, 8,296 chat messages were collected and analyzed using automated text analysis. Our results showed that, based on self-identified gender, men streamers’ viewers employed more swear words than women streamers’ viewers. Meanwhile, women streamers’ viewers used more polite words than men streamers’ viewers. Additionally, we found that the use of both polite and swear words differs significantly between viewers of cisgender and non-binary streamers, which underscores the need for researchers to collect self-identified gender information in social media research
Machine Heuristic at Work: User Evaluations and Folk Theories of Weather News in ‘Immersive Mixed Reality’ Video
No full textImmersive mixed reality (IMR) technologies are transforming news reporting by enhancing audience engagement and improving the visualization of meteorological phenomena. While these technologies offer benefits for communicating public issues through news media, their complexity may pose challenges to audience understanding and trust in the information presented. This study adopted a mixed-methods approach to investigate how users evaluate news videos produced using IMR techniques compared to the traditional text-based format, and how individual differences in machine heuristics influence evaluations and folk theories. The immersive format is more effective in enhancing users’ sense of novelty and personal relevance. Machine heuristic significantly positively moderates the effect of news format on personal relevance. While a high machine heuristic is associated with positive perceptions and the data-driven nature of technology, low machine heuristic individuals form intuitive theories that emphasize AI involvement. Implications for the use and disclosure of immersive technology in news are discussed
Mining Customer Journeys to Uncover Empirical Retail Agglomerations
Get PDFShopping centers are a cornerstone of the retail system, with their success hinging on offering tenant mixes and layouts that stimulate cross-shopping. Despite the rise of e-commerce, physical retail remains vital as consumers increasingly seek blended digital–in-store experiences. Yet, traditional approaches to analyzing shopper behavior often rely on surveys or simple frequency counts, which fail to capture the complexity of customer journeys. This study addresses this gap by applying spatial big data and unsupervised machine learning to investigate empirical retail agglomerations. The research explores how structured, non-random co-visitation patterns within a shopping center can be systematically identified and leveraged within a tenant-mix strategy. Drawing on 24 million anonymized visits to a major Canadian shopping center, the study employs GeoAI and association rule mining, specifically the Apriori algorithm, to uncover high-frequency and high-lift co-visitation rules. Results reveal structured journeys that highlight strong co-visitation between anchors and specialty tenants, confirming that shopping center behavior is far from random. These patterns suggest optimal adjacencies and provide a data-driven framework for leasing and tenant layout. The study contributes theoretically by extending retail agglomeration research using unsupervised methods to examine behavioral clustering on a large-scale dataset empirically. For practitioners, the research approach offers actionable insights for leasing and tenant-mix optimization
Reducing Cognitive Biases in Business Intelligence: A Framework for Objective Data Analysis
Get PDFThis paper examines the ways in which cognitive biases affect judgment in business intelligence (BI) settings. Drawing on bounded rationality theory, the study explores how human cognition, BI system design, and organizational pressure interact to produce biases such as anchoring bias and confirmation bias. Qualitative data were obtained from semi-structured interviews with representatives from two logistics companies. The results of a thematic analysis revealed that users frequently rely on heuristics because of time pressure, information overload, and visual design elements. While some informal bias-mitigation strategies exist, institutional support for them is limited. This study offers a framework for understanding and reducing cognitive bias in BI use, with implications for dashboard design, user training, and decision processes
AI, Uncertainty and Unethical Pro-Organizational Behavior (UPOB)
Get PDFUncertainty, unlike risk, cannot be analyzed using informed probabilities for alternative outcomes. Many pre-deployment technology outcomes are inherently uncertain, as is the case with generative AI. Such innovative developments are non-monotonic due to uncertainty, requiring design reconsiderations as new information emerges. Development teams must nevertheless weigh uncertain outcomes when deciding how to proceed. This paper introduces the concept of ‘Uncertain UPOB’ (U2POB) to distinguish behaviors where the unethical outcome is not guaranteed but remains possible due to uncertainty. Data was collected via a vignette-based survey of 101 technology professionals (606 decisions), with UPOB support found to be 22.1% higher under uncertain the conditions. Overall 35% of respondents were found to be supportive of at least one UPOB decision. Regression analyses indicated that subject UPOB propensity and personal responsibility psychometrics were predictive of UPOB behaviors in both the certain and uncertain contexts
Characterizing YouTube Channels via Multi-View Content Similarity
Get PDFYouTube channels communicate their themes and attract viewers through titles, descriptions, transcripts, and categories. While prior research has focused on user engagement, the internal consistency of content across a channel’s videos remains underexplored. This paper presents a framework for characterizing YouTube channels based on semantic similarity among key content features. Using a dataset of 150 channels and 157,235 videos, pairwise similarity scores were computed across six content feature combinations. five unsupervised clustering algorithms were applied to each pair, and results were integrated through majority voting to produce stable channel clusters. Similarity-based analysis revealed recurring alignment patterns, leading to five high-level content behavior characterizations. These profiles reflect distinct strategies in metadata coherence, narrative structure, and category usage. The proposed method enables scalable, label-free, and language-agnostic analysis of channel behavior. By revealing gaps between content and metadata, the framework surfaces potential editorial biases that may disproportionately affect marginalized audiences on social media
A Domain-Adaptive Soft Prompting Framework for Multi-Type Bias Detection in News
Get PDFAdvances in Large Language Models (LLMs) have enabled new opportunities to automate media analysis and improve collaborative social cybersecurity. A key task is bias detection in news reporting, which is essential for promoting information fairness and reducing polarization. However, existing approaches often rely on supervised fine-tuning with labeled datasets and fail to capture domain-specific linguistic patterns, limiting scalability and generalization. To address this, we propose a lightweight, modular framework that combines domain-adaptive pretraining (DAP) with Masked Language Modeling (MLM) and soft prompt tuning to detect six types of media bias (framing, group, semantic properties, connotation, informational spin, and phrasing). Our framework leverages 401,000+ New York Times articles from 2000 to 2024 to pretrain five LLMs, followed by bias prompting with small labeled data. The approach improves F1 by 7.6% and precision by 6.8% over hard prompts on average across the six types of biases. These results confirm DAP with soft prompts as an efficient and scalable solution for bias-aware NLP in resource-constrained environment
From AI Literacy to AI Use: Evidence from a Multi-Organization Upskilling Program
Get PDFAs artificial intelligence (AI) becomes integral to organizational processes, employee adoption emerges as a key determinant of a successful implementation. This study draws on the Unified Theory of Acceptance and Use of Technology (UTAUT) and extends it with AI-specific drivers, AI literacy and attitude toward AI, alongside established factors such as subjective norm and effort expectancy, while exploring the moderating roles of the demographic factors age and gender. We surveyed 180 employees from 39 organizations in Bavaria, Germany, before and after a five-month AI upskilling program. Results show that AI literacy and subjective norm were the strongest predictors of AI use. Age moderated the literacy-use relationship, with diminished effects for older employees. Pre-post training comparisons showed significant gains in AI literacy, AI use, and AI-supported task share. These findings refine UTAUT by incorporating AI literacy and offer practical insights for designing targeted, age-sensitive interventions to drive workplace AI usage
Collaborative LLM Agents for C4 Software Architecture Design Automation
Get PDFSoftware architecture design is a fundamental part of creating every software system. Despite its importance, producing a C4 software architecture model — the preferred notation for such architecture — remains manual and time-consuming. We introduce an LLM-based multi-agent system that automates this task by simulating a dialogue between role-specific experts who analyze requirements and generate the Context, Container, and Component views of the C4 model. Quality is assessed with a hybrid evaluation framework: deterministic checks for structural and syntactic integrity and C4 rule consistency, plus semantic and qualitative scoring via an LLM-as-a-Judge approach. Tested on five canonical system briefs, the workflow demonstrates fast C4 model creation, sustains high compilation success, and delivers semantic fidelity. A comparison of four state-of-the-art LLMs shows different strengths relevant to architectural design. This study contributes to automated software architecture design and its evaluation methods
