As social media increasingly function as platforms for information communicating, they have become key venues for health information seeking while also enabling the rapid diffusion of misinformation. This study examines how linguistic characteristics of misinformation and information consistency between original posts and User-Generated Content (UGC) shape misinformation diffusion, with influencer status considered as a moderating factor. Using a hybrid analytical approach that combines text mining and negative binomial regression, we analyze 10,651 misinformation posts and 615,562 related UGC items concerning winter influenza and vaccines on Rednote. Consequently, we reveal that syntactic linguistic features, including sentence length, word composition, and emoji usage, are significantly associated with all three diffusion-related engagement behaviors, i.e., likes, comments, and reposts. Emotional cues exhibit heterogeneous effects across engagement types, where anxiety and anger suppress diffusion-related behaviors while sadness significantly increases reposting. Interestingly, users are more likely to comment on misinformation when UGC is topically similar to the original post, rather than simply liking or reposting it, suggesting evaluative engagement rather than passive endorsement. Further, at the pragmatic level, metaphorical expression and information consistency across syntactic, semantic, and pragmatic dimensions between source content and UGC significantly amplify misinformation diffusion. Importantly, influencer status positively moderates these relationships, strengthening the effects of linguistic features and information consistency, particularly for reposting. Overall, this study advances misinformation diffusion research by integrating linguistic, cross-text, and social influence mechanisms, and provides actionable insights for platform governance aimed at balancing user engagement with information integrity in digital information platforms.

With the rapid development of Internet technology, information overload has become increasingly severe. Recommendation systems, as effective information filtering tools, can provide users with personalized recommendations to reduce their decision-making costs. Although recommendation algorithms have achieved great progress in prediction accuracy with the advance of deep learning technologies, the reliability of recommendation results has been relatively underexplored. "Reliability" refers to the likelihood of users accepting the final recommended results, and it is one of the important aspects influencing user experience. Many existing recommendation systems primarily optimize predictive accuracy metrics, while relatively fewer works explicitly model the likelihood of user acceptance, leading to potential gaps in user satisfaction. To address this, we propose the deep ensemble reliable recommendation algorithm (DERA). DERA integrates reliability into both the data preprocessing and model training phases of recommendation models. Drawing on the ensemble learning concept, DERA trains multiple weak learners and employs voting to determine the final prediction result. Additionally, a data preprocessing method is designed to alleviate the imbalance of training data. Experiments on four real-world datasets demonstrate that DERA can enhance recommendation performance by considering reliability. In summary, our work presents a novel framework to integrate reliability estimation into the training pipeline without extra side-information.

Large language models (LLMs) are increasingly utilized for named entity recognition (NER) in health care, with significant potential to enhance symptom detection within electronic health records (EHRs). This study explores the application of LLMs to identify symptoms of anxiety and nausea/vomiting documented in the clinical notes of patients with cancer. We analyzed clinical notes from 8,490 patients diagnosed with various cancer types. Bio Clinical BERT and Bio GPT models were further pretrained on clinical text from this dataset. Two modeling strategies, fine-tuning and prompt-based learning, were implemented using Symptom-BERT and Symptom-GPT frameworks. Model performance was evaluated using F1 scores, emphasizing recognizing psychological symptoms (anxiety) and physical symptoms (nausea/vomiting). Fine-tuning with Symptom-BERT achieved the highest F1 scores, 0.989 for nausea/vomiting and 0.912 for anxiety, significantly outperforming Symptom-GPT in detection accuracy. While prompt-based learning with Symptom-GPT surpassed that of a few-shot learning, it remained less effective than fine-tuning. Fine-tuning excelled in identifying well-documented symptoms, particularly physical ones like nausea/vomiting. Using named entity recognition (NER), the study analyzed the entire dataset, detecting anxiety in 2,436 patients (28.69%) and nausea/vomiting in 3,338 patients (39.31%). While both fine-tuning and prompt-based learning approaches offer utility, fine-tuning demonstrates superior accuracy in recognizing symptoms from clinical narratives, particularly physical ones. LLM-based symptom detection can support oncology nurses and care teams by enabling earlier recognition of patient-reported symptoms documented in narrative notes. These tools offer practical value in improving symptom monitoring, care planning, and timely intervention, thereby enhancing patient-centered care in oncology settings.

While prior financial distress prediction (FDP) studies have increasingly incorporated multi-source data, existing approaches rarely integrate financial indicators with fine-grained narrative signals in a unified framework. To bridge this gap, we propose a novel FDP framework that integrates simultaneously financial ratios, Management Discussion and Analysis (MD&A), and investor comments from social media. We first propose an LLM-BERT based triplet extraction approach to capture aspect-level semantic and sentiment information from MD&A texts. Second, we utilize FinBERT to extract sentiment features from investor comments. These textual features are then combined with financial ratios and integrated into a Multi-source Financial Information Fusion Network (MFIFN), trained with focal loss to mitigate class imbalance. Based on the dataset of 24,429 firm-year samples from Chinese listed companies between 2014 and 2023 (including both distressed and non-distressed firms), experimental results demonstrate that incorporating social media and MD&A features provides incremental predictive values on top of financial ratios. In particular, the proposed MFIFN model achieves an AUC of 0.9541. Furthermore, the LLM-BERT based triplet extraction method improves feature quality, delivering consistent performance gains across compared with traditional textual feature extraction methods. These findings suggest that integrating financial indicators with fine-grained narrative signals can enhance early warning systems, providing stakeholders with more timely and comprehensive risk assessment tools.

Online Q&A communities are built around both knowledge sharing and visible interactions among contributors, creating conditions in which contributors may be influenced by the emotional tone of prior answers. Drawing on emotional contagion theory and cognitive reappraisal, this study examines whether and how emotions spread among knowledge contributors, and how such contagion is shaped by cognitive engagement and social status. Using a large-scale dataset from Zhihu, we measure answer sentiment through LIWC-based text analysis, control for topic heterogeneity with topic modeling, and estimate cross-classified hierarchical linear models that account for answers nested within both questions and contributors. We also use an instrumental variable approach to address potential endogeneity. The results show that prior answers’ emotional tone significantly predicts the sentiment of subsequent contributions, indicating emotional contagion in a weak-tie, task-oriented knowledge community. This effect is weakened when contributors engage in more substantive elaboration, suggesting a rationality cooling mechanism, especially for negative sentiment. Further valence-specific analyses reveal an asymmetric role of social status. High-status contributors are less responsive to positive peer sentiment, consistent with an objective image-maintenance motive, but are more responsive to negative peer sentiment, suggesting the strategic use of negativity to signal expertise and critical discernment. This study reframes sentiment in UGC as a dynamic behavioral outcome shaped by peer exposure, cognitive regulation, and status-based self-presentation, offering implications for platform designs that reduce emotional polarization by encouraging cognitive effort.

Executives' online personal brands are regarded as important intangible assets. This research conceptualizes executives' online personal brands as a form of corporate advertising. Drawing on principles from advertising effectiveness and cognitive balance theory, we examine their influence on corporate performance. Using data from 3,702 Chinese A-share listed companies, our empirical analysis reveals that executives' online personal brands positively affect corporate performance. Furthermore, we find that brand visibility and brand attention, as key dimensions of online personal brands, independently contribute to enhanced corporate performance, with corporate reputation as a crucial mediator. Propensity score matching is employed to address endogeneity concerns. This research enriches digital branding literature by exploring the strategic role of executives' online personal brands in driving corporate success. It offers actionable insights for executives and firms seeking to leverage personal branding for corporate success in the digital age.

IntroductionUser-generated video performance underpins the sustainable development of the user-generated video economy, yet its influencing factors remain underexplored. Drawing on the componential theory of creativity and information overload theory, this study examines how user feedback correlates with video performance and whether creator experience moderates such relationships.MethodsEmpirical analyses were performed based on Bilibili user-generated video data. We classified user feedback by quantity and type and adopted regression models to test the hypothesized moderating effects.ResultsUser feedback volume exhibited a significant inverted U-shaped association with video performance. Creator experience positively moderated the relationship’s turning point, elevating the critical feedback volume threshold. Additionally, different feedback types showed heterogeneous correlations with video performance.DiscussionThis study refines the nonlinear mechanism and boundary condition of audience feedback affecting creators’ creative output, enriching relevant theoretical research. It provides differentiated operational and managerial implications for video platforms to optimize content ecology and sustainable development.

Delivery time is one of the keys to e-commerce success. Although retailers are taking steps to offer fast delivery services, these approaches are expensive, and they may not be sustainable in the long term. By leveraging a unique shift resulting from user-interface designs on eBay.com, we find that introducing a nudging-based function (i.e., a guaranteed delivery toggle) is a promising cost-effective approach that may ameliorate the relationship between slow delivery time and product sales. Specifically, our research devises three complementary studies—(1) examining the sales impact of the toggle function using a large-scale secondary data set, (2) investigating the underlying mechanism of the toggle effectiveness through laboratory and online experiments, and (3) analyzing the far-reaching impacts on consumers’ subsequent rating behaviors—to establish the external and internal validity and the robustness of this finding. We attribute the expansion of slow-delivery item sales (Study 1) to the subtle change in the design of the choice architecture from simultaneous attribute choices to sequential attribute choices (Study 2). Study 3 reveals the mitigation of dimensional rating bias toward delivery time, further interpreting the nudging effect of sequential attribute choices. Given the dilemma of e-commerce for fast-delivery provision, our research elucidates how practitioners improve their operational decisions to address the last-mile logistics problem by factoring in a simple and free delivery-related nudge. This paper was accepted by D. J. Wu, information systems. Funding: This research was supported by the National Natural Science Foundation of China [Grants 72101104, 72471106, and 72571259], and it was partially supported by the Soft Science Special Project of the Gansu Basic Research Plan [Grant 24JRZA032] and the Fundamental Research Funds for the Central Universities [Grant 2024lzujbkyqm009]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.01572 .

The rapid rise of short-form video commerce has posed new challenges for accurate sales prediction. Previous studies often overlook the sequential features and temporal structure of modality-specific data. Our main contribution is to develop a signal-driven multimodal sequence fusion (SMSF) framework grounded in signaling theory. SMSF employs a hierarchical attention-based fusion strategy that combines self-attention, pairwise crossmodal attention, and joint attention to enhance sequential modeling and enable effective representation of complex and heterogeneous multimodal signals beyond conventional fusion methods. All enhanced signals are fused via dynamic gating for adaptive modality selection, followed by a Mixture-of-Experts module that improves semantic coordination. We use a multimodal dataset of 8084 short-form sales videos from Douyin (TikTok in China) for experiments. The results indicate that SMSF consistently outperforms benchmark models. Ablation experiments further reveal that online comments contribute most to prediction performance. Its signal processing and enhancement methods provide a potential paradigm for multimodal tasks across different scenarios or platforms.

Open collaboration has emerged as a powerful tool for harnessing collective intelligence to solve complex problems in the rapidly evolving digital era. As organizations increasingly rely on online contests to foster open collaboration, understanding the dynamics of teamwork in open collaboration contests becomes crucial. This study examines how team composition attributes moderate the curvilinear relationship between team size and online team performance in open collaboration contests. Drawing on the benefit-cost framework from utility maximization theory, the results demonstrate that team size has an inverted U-shaped effect on team performance through the tradeoff of expertise aggregation benefits and resource coordination costs. Notably, diversity in team abilities shifts the turning point of this U-shaped curve to the right due to the benefits of expertise aggregation. The presence of star members in a team shifts the turning point of this U-shaped curve to the left due to the additional costs of exponential growth in resource coordination. Our work makes contributions to the team management literature by uncovering the dynamic associations between team size, team composition and online team performance in the complex open collaboration environment. Moreover, it offers practical implications for the management of individual contestants and the governance of online contest platforms.

Online reviews play an important role in explaining product demand. Our study extends the literature by assessing how arousal (the intensity of the emotion), an important dimension of emotion, and valence (the positiveness of the emotion) can affect product demand across different customer groups. Specifically, regarding product demand from new customers, consumers infer service quality from expressed valence and arousal from other consumers' reviews. Regarding product demand from returning customers, consumers develop perceptions of service quality based on experienced valence and arousal (reflected in their reviews). We then examine product demand from new customers (studies 1 and 2a) as well as that from returning customers (studies 1 and 2b), and establish the external and internal validity regarding the moderating effects of arousal on different types of product demand. The results reveal that, although arousal has positive moderating effects for both new and returning customers, the strengths of the moderating effects differ. Furthermore, valence and arousal from other customers do not have significant effects for returning customers. Our study contributes to the literature by clarifying the moderating effect of arousal on product demand and by providing a more nuanced understanding of product demand at the customer profile level. Our results thus offer important practical suggestions regarding how to explain product demand across different customer groups by considering the textual information (i.e., arousal) of customer reviews.

Patients are increasingly turning to online health Q A communities for social support to improve their well-being. However, when this support received does not align with their specific needs, it may prove ineffective or even detrimental. This necessitates a model capable of identifying the social support needs in questions. However, training such a model is challenging due to the scarcity and class imbalance issues of labeled data. To overcome these challenges, we follow the computational design science paradigm to develop a novel framework, Hybrid Approach for SOcial Support need classification (HA-SOS). HA-SOS integrates an answer-enhanced semi-supervised learning approach, a text data augmentation technique leveraging large language models (LLMs) with reliability- and diversity-aware sample selection mechanism, and a unified training process to automatically label social support needs in questions. Extensive empirical evaluations demonstrate that HA-SOS significantly outperforms existing question classification models and alternative semi-supervised learning approaches. This research contributes to the literature on social support, question classification, semi-supervised learning, and text data augmentation. In practice, our HA-SOS framework facilitates online Q A platform managers and answerers to better understand users' social support needs, enabling them to provide timely, personalized answers and interventions.

It is known that streamers play a special role in live streaming commerce, but there is a huge discrepancy in sales performance resulting from different characteristics of streamers. This study applies social influence theory to systematically analyze how streamer characteristics interact to affect sales performance. Using a unique dataset of 120,794 live streaming records from 597 streamers on Douyin platform, we establish a fixed effects model with unbalanced panel data. The results show that previous sales have strong momentum effects. Total views, number of live commercial products and live streaming duration all have a positive impact on sales volumes. Heterogeneity analysis reveals significant differences across identity types, industry types, and authentication statuses, with celebrities, streamers from entertainment and leisure sectors, and unverified streamers showing notably stronger gains. These findings provide empirical evidence to guide streamers and platforms in optimizing marketing strategies in the competitive live streaming commerce.

Session-based recommendation aims to provide recommendation list based on the recent interaction behaviors of anonymous users. Existing session-based recommender systems adopt various methods to capture the behavioral patterns of anonymous users in the current session. However, most of the recommenders assumed that the type of positional information is single, neglecting the influence of other potential positional information on learning preferences. Although a few related works consider the impact of multiple positional information, they are limited in performance by inefficient fusion methods. These methods mainly based on invasive fusion which mixes different types of positional information together, fails to distinguish the various preferences and overwhelms the item representation. To address these challenges, we propose Coformer-DP which fuses user preferences hidden in general and personalized positional information based on the non-invasive method. Furthermore, the adopted contextual preference efficiently regulates the long-term preference and the comprehensive preference. Extensive experiments on four real-world datasets show the distinguished performance of Coformer-DP.

Session-based recommendation systems (SBRSs) predict the next item in a session by analyzing user interactions. While current methods emphasize sequential item relationships, they often overlook temporal information that highlights subtle shifts in user preferences. This gap can limit their ability to adapt to dynamic user behavior, and recent advances have yet to effectively integrate both sequential and non-sequential item transitions, which may lead to biased modeling. To address these limitations, this paper introduces Coase, a novel SBRS model that unifies local and global context modeling to capture fine-grained dynamic user preferences. Coase transforms session sequences into session star graphs, employing a Bi-Gated Graph Self-Attention Network for local context modeling, and introduces SudokuFormer to model time-aware sequential transitions within a global session context through disentangled attention and stable feature fusion. A triple attention mechanism is then utilized to fully integrate local and global contextual features. Comprehensive experiments conducted on four publicly available datasets demonstrate that Coase improves Recall by 1.71%-1.83%, Mean Reciprocal Rank (MRR) by 2.73%-2.80%, and Normalized Discounted Cumulative Gain (NDCG) by 2.32%-2.43% across the top 5, 10, 15, and 20 items. Ablation studies validate the framework and components of Coase, while additional analyses examine the effect of session length, and visualization studies illustrate diverse attention patterns. This research contributes a novel approach to SBRS, offering promising advancements in recommendation accuracy and user experience.

It is becoming an important marketing strategy for enterprises to utilise user-generated content (UGC) to build brands and promote products on social media platforms, but little is known about the influence mechanism of the popularity of UGC related to brands and products. Drawing upon information adoption theory, this study developed a conceptual model to explain what makes UGC more popular in social media context. Using a dataset of 16,974 records related to the UGC about products from Xiaohongshu app, the most popular user experience sharing platform in China, the empirical study indicates that factors extracted from information quality and source credibility have significant effects on UGC popularity. Interestingly, text length and strength of text sentiment have different effects moderated by product types. In addition, media richness may lead to information redundancy and weaken the effect of text length. The findings offer an insightful understanding into the mechanism of utilising UGC in product marketing on social media platforms.

Environmental security, economic development, and improved well-being are the core issues in high-quality development. The Qinghai-Tibet Plateau is relatively rich in resources, but its overall economic level is relatively low and cannot meet the basic requirements for improving human well-being. The coordinated development of “production, life, and ecology” needs to be addressed urgently. In this context, this study conducted an in-depth analysis of the impact of economic development and environmental security on human well-being in 17 prefecture-level cities on the Qinghai-Tibet Plateau. First, we selected 22 indicators to measure human well-being and used principal component analysis to synthesize each indicator into five dimensions: income and consumption (IAC), means of production (MOP), means of subsistence (MOS), resource acquisition ability (RAA), and physical health (IGH). Second, it analyzes the trend of economic development and environmental security indicators in 17 cities in the Qinghai-Tibet Plateau from 2007 to 2018 and uses a dynamic panel model to analyze the impact of economic development and environmental security on human well-being. The results show that economic development promotes human well-being in the IAC, MOS, and IGH dimensions, whereas environmental security inhibits human well-being in the IAC, RAA, and IGH dimensions.

Despite tremendous recent progress, extant artificial intelligence (AI) still falls short of matching human learning in effectiveness and efficiency. One fundamental disparity is that humans possess a wealth of prior knowledge, while AI lacks the essential commonsense knowledge required for learning tasks. Guided by schema theory, we employ the design science research methodology to introduce a novel knowledge-aware learning framework to harness the knowledge-based processes in human learning. Unlike existing pre-trained large language models (LLMs) and knowledge-aware approaches that treat knowledge in considerably different ways from humans, our theoretically grounded framework closely mimics how humans acquire, represent, activate, and utilize knowledge. The extensive evaluations in the context of text analytics tasks demonstrate that our design achieves comparable performance to the state-of-the-art LLMs and enhances model generalizability and learning efficiency. This study takes a step forward by bringing cognitive science into building cognitively plausible AI and human-AI collaboration research.

A healthy rural ecosystem ensures a win–win situation for both economic growth and ecological conservation. However, the impact of land use changes at the rural level on ecosystem health remains unclear. This study focuses on the rural scale of Zheng–Bian–Luo, analyzing changes in land use from 2000 to 2020. Using the “Ecosystem Vigor-Organization-Resilience-Services” model, the study evaluates the spatiotemporal patterns of ecosystem health. The Patch-generating Land Use Simulation (PLUS) model was employed to simulate land use and ecosystem health in 2035 under three scenarios: Natural Development (ND), Ecological Protection (EP), and Cropland Protection (CP). The findings are as follows: (1) From 2000 to 2020, the area of cultivated land in Zheng–Bian–Luo rural areas decreased, and the area of forest land first decreased and then increased. (2) During the study period, ecosystem health improved as ecosystem vigor, organization, and services increased. Low-value areas of ecosystem health showed a shrinking trend, most notably in Kaifeng. (3) By 2035, under the EP scenario, forest land increased by 76.794 km2, while it decreased under the CP and ND scenarios. Construction land showed an increasing trend in all three scenarios, with the ND scenario seeing the largest increase of 718.007 km2. (4) In 2035, ecosystem health is projected to decline under the ND scenario due to reduced forest land and increased construction land. The CP scenario showed no significant change in ecosystem health, but the southwestern rural areas of Luoyang improved. The EP scenario saw an overall increase in ecosystem health, highlighting land use optimization as beneficial. Local governments are encouraged to create ecological protection plans balancing ecological and cultivated land protection, focusing on sensitive areas such as the Songshan region and southwestern mountainous areas of Luoyang for coordinated development.

Understanding the resilience capabilities of restaurant operations and the determinants affecting these capabilities is critical to helping restaurants overcome the hardships owing to the coronavirus disease (COVID-19) pandemic. This article adopts a textual analytics approach to scientifically measure consumption trends and identify the shock to restaurant sales using online customer review data from Dianping.com (an O2O platform in China). Moreover, the article proposes a theoretical model of business resilience for the restaurant industry in the context of the pandemic. Then, an empirical investigation on how the determinants in our theoretical framework affect the resilience of restaurant business operations using the panel logit model is conducted. Our findings indicate that the pandemic has severely disrupted the full-service restaurants as compared to the quick-service restaurants. We identify four determinants of resilience, namely social capital (i.e., restaurant rating), physical capital (i.e., contactless service), economic capital (i.e., chain operation), and natural capital (e.g., location), which are significantly associated with the resilience of restaurant business during the pandemic. These four determinants play different roles in the resilience of full-service and quick-service restaurants. The findings of this study have theoretical contribution and generate some important managerial implications for helping the restaurant industry recover from disruptions brought by the COVID-19 pandemic.