Decision Support Systems最新文献_第2页

How does artificial intelligence-generated content reshape user-generated content? An empirical study from TripAdvisor 人工智能生成内容（AIGC）如何重塑用户生成内容（UGC）？来自TripAdvisor的一项实证研究

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-16 DOI: 10.1016/j.dss.2026.114623

Zidong Li, Youngsok Bang

Generative AI has revolutionized content creation across digital ecosystems, yet its broader implications for user-generated content (UGC) remain underexplored. To address this gap, we examine TripAdvisor's introduction of AI-Generated Content (AIGC) summaries and investigate how this feature influences user review behavior. Drawing on a taxonomy of online review-writing motivations, we propose that AIGC fulfills multiple motivations to share experiences, reducing users' incentives to contribute new content. Utilizing a natural experiment with hotel reviews in Singapore, our difference-in-differences analysis reveals that overall review volume declines significantly after AIGC implementation, with high-rated reviews exhibiting a sharper decrease than low-rated ones. This effect is more pronounced for lower-tier hotels than higher-tier hotels. We also observe that reviewers compose longer reviews and assign slightly lower ratings post-AIGC. Our structural topic modeling also reveals a significant shift in review content from general to specific topics. These findings demonstrate how generative AI reshapes UGC dynamics and highlight practical considerations for platform managers seeking to leverage AI innovation while maintaining the authenticity and diversity of user feedback.

生成式人工智能已经彻底改变了整个数字生态系统的内容创作，但其对用户生成内容（UGC）的更广泛影响仍未得到充分探索。为了解决这一差距，我们研究了TripAdvisor引入的ai生成内容（AIGC）摘要，并研究了这一功能如何影响用户的评论行为。根据在线评论写作动机的分类，我们提出AIGC满足了分享体验的多重动机，减少了用户贡献新内容的动机。利用新加坡酒店评论的自然实验，我们的差异中差异分析显示，在AIGC实施后，总体评论量显著下降，高评价的评论比低评价的评论下降幅度更大。这种影响在低级别酒店中比在高级别酒店中更为明显。我们还观察到，审稿人撰写的审稿时间更长，并且在aigc之后分配的评分略低。我们的结构化主题建模还揭示了复习内容从一般主题到特定主题的重大转变。这些发现展示了生成式人工智能如何重塑UGC动态，并强调了平台管理者在寻求利用人工智能创新的同时保持用户反馈的真实性和多样性的实际考虑。

{"title":"How does artificial intelligence-generated content reshape user-generated content? An empirical study from TripAdvisor","authors":"Zidong Li, Youngsok Bang","doi":"10.1016/j.dss.2026.114623","DOIUrl":"10.1016/j.dss.2026.114623","url":null,"abstract":"<div><div>Generative AI has revolutionized content creation across digital ecosystems, yet its broader implications for user-generated content (UGC) remain underexplored. To address this gap, we examine TripAdvisor's introduction of AI-Generated Content (AIGC) summaries and investigate how this feature influences user review behavior. Drawing on a taxonomy of online review-writing motivations, we propose that AIGC fulfills multiple motivations to share experiences, reducing users' incentives to contribute new content. Utilizing a natural experiment with hotel reviews in Singapore, our difference-in-differences analysis reveals that overall review volume declines significantly after AIGC implementation, with high-rated reviews exhibiting a sharper decrease than low-rated ones. This effect is more pronounced for lower-tier hotels than higher-tier hotels. We also observe that reviewers compose longer reviews and assign slightly lower ratings post-AIGC. Our structural topic modeling also reveals a significant shift in review content from general to specific topics. These findings demonstrate how generative AI reshapes UGC dynamics and highlight practical considerations for platform managers seeking to leverage AI innovation while maintaining the authenticity and diversity of user feedback.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"203 ","pages":"Article 114623"},"PeriodicalIF":6.8,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995241","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

MediHC: An AI-powered framework for hierarchical disease classification using multi-head attention and contrastive learning MediHC：使用多头注意和对比学习进行分层疾病分类的人工智能框架

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-13 DOI: 10.1016/j.dss.2025.114592

Yechi Xu , Shaokun Fan , Hongxun Jiang

Timely and accurate diagnosis of complex diseases, particularly those with atypical symptoms, is crucial for reducing patient suffering and healthcare costs. However, current early-stage diagnostic accuracy is often compromised by unstructured patient narratives and limited clinical exposure to rare cases. To address these challenges, we introduce the Medical Hierarchy Classifier (MediHC), an AI-powered framework designed to enhance clinical decision-making by analyzing patient–doctor conversations. MediHC comprises three novel modules: a Language Processing Module using large language models (LLMs) and BioClinicalBERT for medical text embeddings; a Hierarchical Multi-Head Attention Module for modeling disease taxonomy dependencies; and a Multi-Level Prediction Module with Contrastive Learning to distinguish between similar diseases. A composite loss function jointly optimizes predictive performance and feature quality. Extensive experiments on real-world clinical datasets demonstrate that MediHC significantly outperforms existing methods, achieving superior accuracy across multiple levels of disease classification. These results underscore MediHC’s substantial potential to advance diagnostic strategies in challenging clinical contexts.

及时和准确地诊断复杂疾病，特别是那些具有非典型症状的疾病，对于减少患者痛苦和医疗保健费用至关重要。然而，目前早期诊断的准确性经常受到不结构化的患者叙述和对罕见病例的有限临床接触的影响。为了应对这些挑战，我们引入了医疗层次分类器（MediHC），这是一个基于人工智能的框架，旨在通过分析患者与医生的对话来增强临床决策。MediHC包括三个新颖的模块：使用大型语言模型（LLMs）的语言处理模块和用于医学文本嵌入的BioClinicalBERT；基于多级关注模块的疾病分类依赖关系建模以及具有对比学习的多级预测模块，以区分类似疾病。复合损失函数可以同时优化预测性能和特征质量。在真实世界的临床数据集上进行的大量实验表明，MediHC显著优于现有的方法，在多个级别的疾病分类中实现了卓越的准确性。这些结果强调了MediHC在具有挑战性的临床环境中推进诊断策略的巨大潜力。

{"title":"MediHC: An AI-powered framework for hierarchical disease classification using multi-head attention and contrastive learning","authors":"Yechi Xu , Shaokun Fan , Hongxun Jiang","doi":"10.1016/j.dss.2025.114592","DOIUrl":"10.1016/j.dss.2025.114592","url":null,"abstract":"<div><div>Timely and accurate diagnosis of complex diseases, particularly those with atypical symptoms, is crucial for reducing patient suffering and healthcare costs. However, current early-stage diagnostic accuracy is often compromised by unstructured patient narratives and limited clinical exposure to rare cases. To address these challenges, we introduce the Medical Hierarchy Classifier (MediHC), an AI-powered framework designed to enhance clinical decision-making by analyzing patient–doctor conversations. MediHC comprises three novel modules: a Language Processing Module using large language models (LLMs) and BioClinicalBERT for medical text embeddings; a Hierarchical Multi-Head Attention Module for modeling disease taxonomy dependencies; and a Multi-Level Prediction Module with Contrastive Learning to distinguish between similar diseases. A composite loss function jointly optimizes predictive performance and feature quality. Extensive experiments on real-world clinical datasets demonstrate that MediHC significantly outperforms existing methods, achieving superior accuracy across multiple levels of disease classification. These results underscore MediHC’s substantial potential to advance diagnostic strategies in challenging clinical contexts.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"203 ","pages":"Article 114592"},"PeriodicalIF":6.8,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145961667","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Should amazon display product Q&As more prominently? The informational role of Q&As and reviews, and the moderating effect of product involvement 亚马逊是否应该更突出地展示产品问答？问答和评审的信息作用，以及产品参与的调节作用

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-09 DOI: 10.1016/j.dss.2026.114613

Gaurav Jetley , Shivendu Shivendu

Amazon recently demoted on-page Q&As by moving them off the main product page and launched Rufus, an AI assistant trained on reviews, Q&As, and catalog data. This redesign foregrounds a core interface question: which information should be surfaced to consumers, and when? We study how the thematic overlap and novelty between the visible “top” reviews and Q&As relate to product sales, and whether these relationships differ by product involvement. Leveraging a large panel dataset of high- and low-involvement products, we use machine learning techniques to quantify the thematic overlap and divergence in content and apply an instrumental variable approach with fixed effects estimator to analyze their impact on sales. Our findings reveal that for high-involvement products, novel information between reviews and Q&As significantly enhances sales by reducing consumer uncertainty. Conversely, for low-involvement products, overlapping information across these sources facilitates purchasing decisions, leading to increased sales. A counterfactual analysis indicates that adding a single, strategically chosen review or Q&A to the visible head can lift sales, especially for low-performing items. We translate these findings into involvement-aware rules for placement and ranking: preserve co-located, complementary Q&As for high-involvement decisions; surface concise cross-source confirmations for low-involvement ones. Our study contributes to the broader understanding of how UGC can be optimized to support consumer decision-making and improve operational effectiveness in e-commerce.

亚马逊最近将页面上的问题从主产品页面上移走，降低了它们的地位，并推出了鲁弗斯（Rufus），这是一款经过评论、问题和目录数据培训的人工智能助手。这种重新设计提出了一个核心的界面问题：哪些信息应该显示给消费者，什么时候显示？我们研究了可见的“热门”评论和Q&；As与产品销售之间的主题重叠和新颖性，以及这些关系是否因产品参与而不同。利用高参与度和低参与度产品的大型面板数据集，我们使用机器学习技术来量化内容中的主题重叠和分歧，并应用固定效应估计器的工具变量方法来分析它们对销售的影响。我们的研究结果表明，对于高介入产品，评论和问答之间的新信息通过减少消费者的不确定性显著提高了销售。相反，对于低参与度的产品，跨这些来源的重叠信息促进了购买决策，从而增加了销售额。一项反事实分析表明，在可见的标题中添加一条策略性的评论或Q&；A可以提高销售额，尤其是对表现不佳的商品。我们将这些发现转化为参与意识规则，用于放置和排名：保留共同定位，互补的问题；表面简洁的跨源确认低介入。我们的研究有助于更广泛地了解如何优化UGC以支持消费者决策和提高电子商务的运营效率。

{"title":"Should amazon display product Q&As more prominently? The informational role of Q&As and reviews, and the moderating effect of product involvement","authors":"Gaurav Jetley , Shivendu Shivendu","doi":"10.1016/j.dss.2026.114613","DOIUrl":"10.1016/j.dss.2026.114613","url":null,"abstract":"<div><div>Amazon recently demoted on-page Q&As by moving them off the main product page and launched Rufus, an AI assistant trained on reviews, Q&As, and catalog data. This redesign foregrounds a core interface question: which information should be surfaced to consumers, and when? We study how the thematic overlap and novelty between the visible “top” reviews and Q&As relate to product sales, and whether these relationships differ by product involvement. Leveraging a large panel dataset of high- and low-involvement products, we use machine learning techniques to quantify the thematic overlap and divergence in content and apply an instrumental variable approach with fixed effects estimator to analyze their impact on sales. Our findings reveal that for high-involvement products, novel information between reviews and Q&As significantly enhances sales by reducing consumer uncertainty. Conversely, for low-involvement products, overlapping information across these sources facilitates purchasing decisions, leading to increased sales. A counterfactual analysis indicates that adding a single, strategically chosen review or Q&A to the visible head can lift sales, especially for low-performing items. We translate these findings into involvement-aware rules for placement and ranking: preserve co-located, complementary Q&As for high-involvement decisions; surface concise cross-source confirmations for low-involvement ones. Our study contributes to the broader understanding of how UGC can be optimized to support consumer decision-making and improve operational effectiveness in e-commerce.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"203 ","pages":"Article 114613"},"PeriodicalIF":6.8,"publicationDate":"2026-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981682","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Financial statement fraud detection using topic-driven financial sentiment analysis 基于主题驱动的财务情绪分析的财务报表舞弊检测

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-08 DOI: 10.1016/j.dss.2026.114615

Petr Hajek , Josef Novotny , Michal Munk

Financial statement fraud undermines market integrity and incurs substantial costs for investors, regulators, and companies. Text-based detection methods have emerged as useful complements to traditional financial indicators, but many fail to incorporate domain-specific topics or sentiment cues, often missing subtle changes in deceptive communication. To overcome this problem, this study proposes a topic-driven financial sentiment analysis (TDFSA) model that detects corporate fraud by analyzing linguistic patterns in the Management Discussion & Analysis (MD&A) sections of annual reports. Our approach captures contextual sentiment within financially relevant topics using FinBERT embeddings. To evaluate these signals in fraud detection, we integrate the TDFSA outputs into a broader cost-sensitive evaluation framework. This framework combines text-based indicators with financial ratios to balance the need to avoid false alarms with the high cost of undetected fraud. Using data from U.S. firms flagged in SEC Accounting and Auditing Enforcement Releases from 2014 to 2024 and matched non-fraud peers, we examine trends in financial ratios, textual complexity, and sentiment dynamics in the three years preceding fraud events. The results show that models leveraging TDFSA achieve higher detection accuracy and lower cost than dictionary-based sentiment, generic topic models, and deep learning baselines.

财务报表欺诈破坏了市场诚信，给投资者、监管机构和公司带来了巨大的成本。基于文本的检测方法已成为传统财务指标的有用补充，但许多方法未能纳入特定领域的主题或情绪线索，往往错过了欺骗性沟通中的微妙变化。为了克服这个问题，本研究提出了一个主题驱动的财务情绪分析（TDFSA）模型，该模型通过分析年度报告中管理层讨论和分析（MD& a）部分中的语言模式来检测企业欺诈。我们的方法使用FinBERT嵌入捕捉金融相关主题中的上下文情感。为了评估欺诈检测中的这些信号，我们将TDFSA的输出整合到一个更广泛的成本敏感评估框架中。该框架将基于文本的指标与财务比率相结合，以平衡避免虚假警报的需要与未被发现的欺诈行为的高成本。利用美国证券交易委员会会计和审计执法发布的2014年至2024年美国公司的数据，以及与非欺诈同行相匹配的数据，我们研究了欺诈事件发生前三年的财务比率、文本复杂性和情绪动态的趋势。结果表明，与基于词典的情感、通用主题模型和深度学习基线相比，利用TDFSA的模型实现了更高的检测精度和更低的成本。

{"title":"Financial statement fraud detection using topic-driven financial sentiment analysis","authors":"Petr Hajek , Josef Novotny , Michal Munk","doi":"10.1016/j.dss.2026.114615","DOIUrl":"10.1016/j.dss.2026.114615","url":null,"abstract":"<div><div>Financial statement fraud undermines market integrity and incurs substantial costs for investors, regulators, and companies. Text-based detection methods have emerged as useful complements to traditional financial indicators, but many fail to incorporate domain-specific topics or sentiment cues, often missing subtle changes in deceptive communication. To overcome this problem, this study proposes a topic-driven financial sentiment analysis (TDFSA) model that detects corporate fraud by analyzing linguistic patterns in the Management Discussion & Analysis (MD&A) sections of annual reports. Our approach captures contextual sentiment within financially relevant topics using FinBERT embeddings. To evaluate these signals in fraud detection, we integrate the TDFSA outputs into a broader cost-sensitive evaluation framework. This framework combines text-based indicators with financial ratios to balance the need to avoid false alarms with the high cost of undetected fraud. Using data from U.S. firms flagged in SEC Accounting and Auditing Enforcement Releases from 2014 to 2024 and matched non-fraud peers, we examine trends in financial ratios, textual complexity, and sentiment dynamics in the three years preceding fraud events. The results show that models leveraging TDFSA achieve higher detection accuracy and lower cost than dictionary-based sentiment, generic topic models, and deep learning baselines.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"203 ","pages":"Article 114615"},"PeriodicalIF":6.8,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145947956","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Corrigendum to “‘Decoding LLMs’ verbal deception in online reviews” [Decision Support Systems 200 (2026) 114529]. “解码法学硕士在线评论中的口头欺骗”的勘误表[决策支持系统200(2026)114529]。

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-07 DOI: 10.1016/j.dss.2025.114594

Yinghui Huang , Jinyi Zhou , Wanghao Dong , Weiqing Li , Maomao Chi , Changbin Jiang , Weijun Wang , Shasha Deng

引用次数: 0

How to leverage digital platforms in enhancing organizational resilience: The roles of supply chain integration and market orientation 如何利用数字平台增强组织弹性：供应链整合和市场导向的作用

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-07 DOI: 10.1016/j.dss.2026.114612

Qinyao Zheng , Jiabao Lin , Jose Benitez

Despite the potential of digital platforms in promoting organizational resilience, the intermediate mechanisms and contextual contingencies of this association remain inadequately explored. Drawing on dynamic capability theory, we investigate how digital platform use influences organizational resilience through supply chain integration (SCI), with market orientation serving as a critical contingency factor. Using a sample of 178 Chinese agribusinesses, we find that both digital platform exploitative use and digital platform explorative use significantly improve SCI, which subsequently enhances organizational resilience. SCI exerts as a partial mediator in the association of digital platform exploitative use with organizational resilience, whereas acts as a full mediator in the association of digital platform explorative use with organizational resilience. Notably, market orientation strengthens the positive association of digital platform exploitative use with SCI, thus amplifying the positive mediating effect of SCI in the association of digital platform exploitative use with organizational resilience. Conversely, market orientation diminishes the favorable influence of digital platform explorative use on SCI, thereby impairing the positive mediating effect of SCI in the association of digital platform explorative use with organizational resilience. This study enriches the IS literature on the business implications of digital platforms by providing theoretical illustrations and empirical evidence on how digital platform use helps agribusinesses develop organizational resilience.

尽管数字平台在促进组织弹性方面具有潜力，但这种关联的中间机制和情境偶然性仍未得到充分探讨。利用动态能力理论，我们研究了数字平台的使用如何通过供应链整合（SCI）影响组织弹性，其中市场导向是一个关键的应急因素。以178家中国农业综合企业为样本，我们发现数字平台的开发性使用和探索性使用都显著提高了SCI，进而提高了组织弹性。SCI在数字平台剥削性使用与组织弹性之间起部分中介作用，在数字平台探索性使用与组织弹性之间起完全中介作用。值得注意的是，市场导向强化了数字平台利用与SCI之间的正相关关系，从而放大了SCI在数字平台利用与组织弹性之间的正向中介作用。相反，市场导向减弱了数字平台探索性使用对SCI的有利影响，从而削弱了SCI在数字平台探索性使用与组织弹性之间的正向中介作用。本研究通过提供关于数字平台使用如何帮助农业企业发展组织弹性的理论说明和实证证据，丰富了关于数字平台商业影响的IS文献。

{"title":"How to leverage digital platforms in enhancing organizational resilience: The roles of supply chain integration and market orientation","authors":"Qinyao Zheng , Jiabao Lin , Jose Benitez","doi":"10.1016/j.dss.2026.114612","DOIUrl":"10.1016/j.dss.2026.114612","url":null,"abstract":"<div><div>Despite the potential of digital platforms in promoting organizational resilience, the intermediate mechanisms and contextual contingencies of this association remain inadequately explored. Drawing on dynamic capability theory, we investigate how digital platform use influences organizational resilience through supply chain integration (SCI), with market orientation serving as a critical contingency factor. Using a sample of 178 Chinese agribusinesses, we find that both digital platform exploitative use and digital platform explorative use significantly improve SCI, which subsequently enhances organizational resilience. SCI exerts as a partial mediator in the association of digital platform exploitative use with organizational resilience, whereas acts as a full mediator in the association of digital platform explorative use with organizational resilience. Notably, market orientation strengthens the positive association of digital platform exploitative use with SCI, thus amplifying the positive mediating effect of SCI in the association of digital platform exploitative use with organizational resilience. Conversely, market orientation diminishes the favorable influence of digital platform explorative use on SCI, thereby impairing the positive mediating effect of SCI in the association of digital platform explorative use with organizational resilience. This study enriches the IS literature on the business implications of digital platforms by providing theoretical illustrations and empirical evidence on how digital platform use helps agribusinesses develop organizational resilience.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"203 ","pages":"Article 114612"},"PeriodicalIF":6.8,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981461","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Learning user preferences in livestreaming market: A graphical model considering temporal effect 直播市场中用户偏好的学习：一个考虑时间效应的图形模型

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-05 DOI: 10.1016/j.dss.2025.114600

Qingyuan Lin , Yijun Li , Miłosz Kadziński , Mengzhuo Guo

The livestreaming market has experienced rapid growth, making effective recommendation systems essential for enhancing user engagement and marketing strategies. Traditional models often fall short in simultaneously capturing user preferences, host popularity, and the temporal dynamics inherent in livestreaming platforms. To address these challenges, we propose an interpretable graphical model that integrates Poisson Factorization with hierarchical structures and explicit temporal effects. Our model jointly learns user preferences and host popularity while accounting for temporal variations. We develop a variational Bayesian inference algorithm for efficient parameter estimation. Using real-world data from a leading livestreaming platform, we demonstrate that our model outperforms several baseline methods in predicting viewing volumes and capturing user–host interactions before, during, and after a public vacation. Additionally, the learned low-dimensional representations enhance predictive tasks, such as payment behavior prediction, and enable effective profiling and segmentation of users and hosts. Our findings provide insights for decision-makers aiming to optimize recommendation systems and marketing strategies in the dynamic livestreaming market.

直播市场经历了快速增长，有效的推荐系统对于提高用户参与度和营销策略至关重要。传统模型在同时捕捉用户偏好、主持人受欢迎程度和直播平台固有的时间动态方面往往存在不足。为了解决这些挑战，我们提出了一个可解释的图形模型，该模型将泊松分解与层次结构和显式时间效应相结合。我们的模型在考虑时间变化的同时，共同学习用户偏好和主机受欢迎程度。我们开发了一种变分贝叶斯推理算法，用于有效的参数估计。使用来自领先直播平台的真实世界数据，我们证明了我们的模型在预测观看量和捕获公共假期之前，期间和之后的用户-主机交互方面优于几种基线方法。此外，学习到的低维表示增强了预测任务，如支付行为预测，并能够有效地分析和分割用户和主机。我们的研究结果为决策者在动态直播市场中优化推荐系统和营销策略提供了见解。

{"title":"Learning user preferences in livestreaming market: A graphical model considering temporal effect","authors":"Qingyuan Lin , Yijun Li , Miłosz Kadziński , Mengzhuo Guo","doi":"10.1016/j.dss.2025.114600","DOIUrl":"10.1016/j.dss.2025.114600","url":null,"abstract":"<div><div>The livestreaming market has experienced rapid growth, making effective recommendation systems essential for enhancing user engagement and marketing strategies. Traditional models often fall short in simultaneously capturing user preferences, host popularity, and the temporal dynamics inherent in livestreaming platforms. To address these challenges, we propose an interpretable graphical model that integrates Poisson Factorization with hierarchical structures and explicit temporal effects. Our model jointly learns user preferences and host popularity while accounting for temporal variations. We develop a variational Bayesian inference algorithm for efficient parameter estimation. Using real-world data from a leading livestreaming platform, we demonstrate that our model outperforms several baseline methods in predicting viewing volumes and capturing user–host interactions before, during, and after a public vacation. Additionally, the learned low-dimensional representations enhance predictive tasks, such as payment behavior prediction, and enable effective profiling and segmentation of users and hosts. Our findings provide insights for decision-makers aiming to optimize recommendation systems and marketing strategies in the dynamic livestreaming market.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"202 ","pages":"Article 114600"},"PeriodicalIF":6.8,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145902325","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Dynamic hypergraph neural networks for consumer purchase path prediction: Integrating promotions, experiences, and store heterogeneity 用于消费者购买路径预测的动态超图神经网络：整合促销、体验和商店异质性

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-05 DOI: 10.1016/j.dss.2026.114614

Juntao Wu , Quan Liu , Lifan Chen , Wenxiang Zhao , Hefu Liu

With the rise of mobile and online food and beverage services, consumer behavior has become increasingly dynamic, multichannel, and personalized. To support improved decision making, restaurant operators need systems that not only predict purchasing behavior but also capture the complex interactions among promotions, consumer experiences, and historical actions. This paper proposes a Behavioral Economics-informed Hyper Graph Network (BEHGN) framework, which integrates Expected Utility Theory and Mental Accounting Theory to model both short-term promotional responses and long-term experience effects. BEHGN employs a hypergraph structure to represent consumers, stores, products, and coupons, and uses large language models to extract experiential features from online reviews. This design enables the system to capture cross-store behavior dynamics and store heterogeneity in a unified decision support model. Experiments on real-world data from two major food and beverage chains demonstrate that BEHGN outperforms existing models in predicting consumer purchase paths, offering higher accuracy and adaptability. The results highlight the potential of BEHGN to enhance food and beverage decision support, contributing to better strategy formulation and performance outcomes.

随着移动和在线餐饮服务的兴起，消费者行为变得越来越动态、多渠道和个性化。为了支持改进的决策制定，餐厅经营者需要的系统不仅要预测购买行为，还要捕捉促销、消费者体验和历史行为之间复杂的相互作用。本文提出了一个基于行为经济学的超图网络（BEHGN）框架，该框架将期望效用理论和心理会计理论结合起来，对短期促销反应和长期体验效应进行建模。BEHGN采用超图结构来表示消费者、商店、产品和优惠券，并使用大型语言模型从在线评论中提取体验特征。这种设计使系统能够在统一的决策支持模型中捕获跨商店行为动态和商店异质性。对两家主要食品和饮料连锁店的真实数据进行的实验表明，BEHGN在预测消费者购买路径方面优于现有模型，具有更高的准确性和适应性。结果突出了BEHGN在提高食品和饮料决策支持方面的潜力，有助于更好的战略制定和绩效结果。

{"title":"Dynamic hypergraph neural networks for consumer purchase path prediction: Integrating promotions, experiences, and store heterogeneity","authors":"Juntao Wu , Quan Liu , Lifan Chen , Wenxiang Zhao , Hefu Liu","doi":"10.1016/j.dss.2026.114614","DOIUrl":"10.1016/j.dss.2026.114614","url":null,"abstract":"<div><div>With the rise of mobile and online food and beverage services, consumer behavior has become increasingly dynamic, multichannel, and personalized. To support improved decision making, restaurant operators need systems that not only predict purchasing behavior but also capture the complex interactions among promotions, consumer experiences, and historical actions. This paper proposes a Behavioral Economics-informed Hyper Graph Network (BEHGN) framework, which integrates Expected Utility Theory and Mental Accounting Theory to model both short-term promotional responses and long-term experience effects. BEHGN employs a hypergraph structure to represent consumers, stores, products, and coupons, and uses large language models to extract experiential features from online reviews. This design enables the system to capture cross-store behavior dynamics and store heterogeneity in a unified decision support model. Experiments on real-world data from two major food and beverage chains demonstrate that BEHGN outperforms existing models in predicting consumer purchase paths, offering higher accuracy and adaptability. The results highlight the potential of BEHGN to enhance food and beverage decision support, contributing to better strategy formulation and performance outcomes.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"202 ","pages":"Article 114614"},"PeriodicalIF":6.8,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145902326","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Alice in land of games: Investigating behavior spillover effects of users' engagement and social connections across games 游戏领域的Alice：调查游戏中用户粘性和社交关系的行为溢出效应

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2026-01-02 DOI: 10.1016/j.dss.2025.114611

Yue Li , Haowen Deng , Cheng Zhang

The rapid growth of digital gaming allows players to move seamlessly between virtual worlds, raising questions about how behaviors in one gaming world transfer to another. Drawing on operant conditioning theory, we posit that engagement—driven by reinforced action–reward loops—forms internalized play habits that spill over positively across games, while social connections create coordination costs that hinder cross-game transitions. Using a panel of 25,164 users observed across two same-type games, we employ user-day-level panel regressions, structural equation modeling, and latent-class analysis to test these hypotheses. Our results reveal three core insights. First, higher engagement in the initial game significantly increases engagement in the subsequent game, whereas stronger social connections in the first game reduce later engagement, confirming positive and negative behavior spillover effects, respectively. Second, both spillover effects are intensified under collaborative play context, where teamwork amplifies reinforcement for highly engaged players but deepens “lock-in” for socially embedded players. Third, engagement-driven users exhibit more consistent cross-game routines, shorter switching intervals, and higher daily switch frequencies than their social-driven counterparts. These findings deepen our understanding of cross-game engagement and offer actionable guidance for optimizing event timing, social features, and retention strategies across game portfolios.

数字游戏的快速发展使玩家能够在虚拟世界之间无缝移动，这引发了一个问题，即一个游戏世界中的行为如何转移到另一个游戏世界。根据操作性条件反射理论，我们认为，在强化的行动奖励循环的驱动下，用户粘性形成了内化的游戏习惯，并在整个游戏中产生积极的影响，而社交联系则产生了阻碍跨游戏过渡的协调成本。通过观察两款同类型游戏的25164名用户，我们使用用户日水平面板回归、结构方程模型和潜在类分析来测试这些假设。我们的研究结果揭示了三个核心见解。首先，在初始游戏中较高的参与度会显著提高后续游戏的参与度，而在初始游戏中较强的社交联系会降低后续游戏的参与度，这分别证实了积极和消极的行为溢出效应。其次，这两种溢出效应在合作游戏情境下都得到强化，团队合作强化了高度投入的玩家，但加深了社交玩家的“锁定”。第三，用户粘性驱动型用户比社交驱动型用户表现出更一致的跨游戏习惯、更短的切换间隔和更高的每日切换频率。这些发现加深了我们对跨游戏粘性的理解，并为优化游戏组合中的活动时间、社交功能和留存策略提供了可行的指导。

{"title":"Alice in land of games: Investigating behavior spillover effects of users' engagement and social connections across games","authors":"Yue Li , Haowen Deng , Cheng Zhang","doi":"10.1016/j.dss.2025.114611","DOIUrl":"10.1016/j.dss.2025.114611","url":null,"abstract":"<div><div>The rapid growth of digital gaming allows players to move seamlessly between virtual worlds, raising questions about how behaviors in one gaming world transfer to another. Drawing on operant conditioning theory, we posit that engagement—driven by reinforced action–reward loops—forms internalized play habits that spill over positively across games, while social connections create coordination costs that hinder cross-game transitions. Using a panel of 25,164 users observed across two same-type games, we employ user-day-level panel regressions, structural equation modeling, and latent-class analysis to test these hypotheses. Our results reveal three core insights. First, higher engagement in the initial game significantly increases engagement in the subsequent game, whereas stronger social connections in the first game reduce later engagement, confirming positive and negative behavior spillover effects, respectively. Second, both spillover effects are intensified under collaborative play context, where teamwork amplifies reinforcement for highly engaged players but deepens “lock-in” for socially embedded players. Third, engagement-driven users exhibit more consistent cross-game routines, shorter switching intervals, and higher daily switch frequencies than their social-driven counterparts. These findings deepen our understanding of cross-game engagement and offer actionable guidance for optimizing event timing, social features, and retention strategies across game portfolios.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"202 ","pages":"Article 114611"},"PeriodicalIF":6.8,"publicationDate":"2026-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145893960","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

What makes a good image? Exploring patients' physician selection behavior leveraging large language models and scenario experiments 是什么造就了一个好的形象？利用大型语言模型和场景实验探索患者的医生选择行为

IF 6.8 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Decision Support Systems

Pub Date : 2025-12-30 DOI: 10.1016/j.dss.2025.114608

Shan Liu , Qingshan Liu , Kezhen Wei , Guangsen Si , Chenze Wang , Muyu Zhang

As important information cues for patients' selection, physicians' online profile images have received limited attention. We explore the effects of visual cues—image feature (image clarity) and image contents (smile intensity and medical professionalism) on patients' selection behavior, while also examining the moderating effect of consultation price. Leveraging large language models, we annotate visual cues to facilitate empirical analysis. This analysis demonstrates that image clarity, smile intensity, and medical professionalism positively affect patients' selection behavior, with consultation price amplifying the effect of image clarity. We further conduct scenario-based experiments to examine the underlying mechanism from perspectives of information foraging and perceived diagnosticity. This study enriches theoretical insights into patients' selection behavior by mining physicians' image information. It also advances the empirical methodological paradigm by integrating the large language model with empirical analysis. Our findings help physicians and platform managers strategically optimize profile images and consultation prices to improve physicians' popularity in online health market.

作为患者选择的重要信息线索，医生的在线个人资料图片受到的关注有限。我们探讨了视觉线索-图像特征（图像清晰度）和图像内容（微笑强度和医疗专业度）对患者选择行为的影响，同时考察了咨询价格的调节作用。利用大型语言模型，我们注释视觉线索以促进实证分析。分析表明，图像清晰度、微笑强度和医疗专业精神正向影响患者的选择行为，且咨询价格放大了图像清晰度的影响。我们进一步从信息觅食和感知诊断的角度进行了基于场景的实验来研究其潜在机制。本研究通过对医生影像信息的挖掘，丰富了对患者选择行为的理论认识。它还通过将大语言模型与实证分析相结合，推进了实证方法论范式。我们的研究结果有助于医生和平台管理者战略性地优化个人资料图像和咨询价格，以提高医生在在线医疗市场的知名度。

{"title":"What makes a good image? Exploring patients' physician selection behavior leveraging large language models and scenario experiments","authors":"Shan Liu , Qingshan Liu , Kezhen Wei , Guangsen Si , Chenze Wang , Muyu Zhang","doi":"10.1016/j.dss.2025.114608","DOIUrl":"10.1016/j.dss.2025.114608","url":null,"abstract":"<div><div>As important information cues for patients' selection, physicians' online profile images have received limited attention. We explore the effects of visual cues—image feature (image clarity) and image contents (smile intensity and medical professionalism) on patients' selection behavior, while also examining the moderating effect of consultation price. Leveraging large language models, we annotate visual cues to facilitate empirical analysis. This analysis demonstrates that image clarity, smile intensity, and medical professionalism positively affect patients' selection behavior, with consultation price amplifying the effect of image clarity. We further conduct scenario-based experiments to examine the underlying mechanism from perspectives of information foraging and perceived diagnosticity. This study enriches theoretical insights into patients' selection behavior by mining physicians' image information. It also advances the empirical methodological paradigm by integrating the large language model with empirical analysis. Our findings help physicians and platform managers strategically optimize profile images and consultation prices to improve physicians' popularity in online health market.</div></div>","PeriodicalId":55181,"journal":{"name":"Decision Support Systems","volume":"202 ","pages":"Article 114608"},"PeriodicalIF":6.8,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145928263","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0