Electronic Commerce Research and Applications最新文献_第2页

A reinforcement learning-driven framework for the Q-commerce multi-product unit scheduling problem Q-commerce多产品单元调度问题的强化学习驱动框架

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2026-01-01 DOI: 10.1016/j.elerap.2025.101571

Weijian Zhang , Min Kong , Weimin Tan , Yingxin Song , Amir M. Fathollahi-Fard

With the rapid emergence of Quick commerce (Q-commerce), e-commerce fulfillment systems are shifting from hour-level to minute-level responsiveness. However, this ultra-fast delivery mode introduces complex operational challenges such as multi-warehouse coordination, multi-stage processing, and high-frequency, small-batch order scheduling. To address these challenges, this study formulates and formally defines the Q-commerce multi-product unit scheduling problem with transportation and setup times, which systematically characterizes the outbound scheduling process under multi-resource collaboration. Given the problem’s strong NP-hard nature, a double Q-Learning-based variable neighborhood search (DQL-VNS) algorithm is developed. This algorithm integrates reinforcement learning with metaheuristic optimization to adaptively select neighborhood operators and adjust perturbation intensity, thereby enabling intelligent self-learning within complex search spaces. Extensive computational experiments show that DQL-VNS effectively reduces makespan and total tardiness. In large-scale instances, it achieves over a 10% reduction in average order delay compared with benchmark algorithms. Moreover, the results reveal that the multi-product unit decomposition strategy significantly enhances outbound efficiency and reduces tardiness. In terms of system configuration, the multi-warehouse-few-station mode outperforms the few-warehouse-multi-station mode by achieving better workload balance and greater fulfillment responsiveness. Additionally, due-date flexibility has a substantial impact on scheduling performance, emphasizing its critical role in maintaining customer satisfaction and delivery reliability. Overall, this study presents a novel modeling perspective and an intelligent optimization framework for outbound scheduling and resource coordination in Q-commerce, providing both theoretical and practical insights for developing responsive and sustainable instant retail logistics systems.

随着快速商务（Q-commerce）的迅速兴起，电子商务履行系统的响应能力正在从小时级向分钟级转变。然而，这种超快速的交付模式带来了复杂的操作挑战，如多仓库协调、多阶段处理和高频、小批量订单调度。为了解决这些问题，本研究提出并正式定义了Q-commerce包含运输和设置时间的多产品单元调度问题，系统地描述了多资源协作下的出库调度过程。针对该问题的强NP-hard特性，提出了一种基于双q学习的变量邻域搜索（DQL-VNS）算法。该算法将强化学习与元启发式优化相结合，自适应地选择邻域算子并调整扰动强度，从而在复杂搜索空间内实现智能自学习。大量的计算实验表明，DQL-VNS有效地降低了完工时间和总延误时间。在大规模实例中，与基准算法相比，它的平均订单延迟减少了10%以上。结果表明，多产品单元分解策略显著提高了出库效率，减少了延误。在系统配置方面，通过实现更好的工作负载平衡和更高的执行响应能力，多仓库-少工作站模式优于少仓库-多工作站模式。此外，截止日期灵活性对调度性能有重大影响，强调其在维持客户满意度和交付可靠性方面的关键作用。总体而言，本研究为Q-commerce的出库调度和资源协调提供了一个全新的建模视角和智能优化框架，为开发响应性和可持续性的即时零售物流系统提供了理论和实践见解。

{"title":"A reinforcement learning-driven framework for the Q-commerce multi-product unit scheduling problem","authors":"Weijian Zhang , Min Kong , Weimin Tan , Yingxin Song , Amir M. Fathollahi-Fard","doi":"10.1016/j.elerap.2025.101571","DOIUrl":"10.1016/j.elerap.2025.101571","url":null,"abstract":"<div><div>With the rapid emergence of Quick commerce (Q-commerce), e-commerce fulfillment systems are shifting from hour-level to minute-level responsiveness. However, this ultra-fast delivery mode introduces complex operational challenges such as multi-warehouse coordination, multi-stage processing, and high-frequency, small-batch order scheduling. To address these challenges, this study formulates and formally defines the Q-commerce multi-product unit scheduling problem with transportation and setup times, which systematically characterizes the outbound scheduling process under multi-resource collaboration. Given the problem’s strong NP-hard nature, a double Q-Learning-based variable neighborhood search (DQL-VNS) algorithm is developed. This algorithm integrates reinforcement learning with metaheuristic optimization to adaptively select neighborhood operators and adjust perturbation intensity, thereby enabling intelligent self-learning within complex search spaces. Extensive computational experiments show that DQL-VNS effectively reduces makespan and total tardiness. In large-scale instances, it achieves over a 10% reduction in average order delay compared with benchmark algorithms. Moreover, the results reveal that the multi-product unit decomposition strategy significantly enhances outbound efficiency and reduces tardiness. In terms of system configuration, the multi-warehouse-few-station mode outperforms the few-warehouse-multi-station mode by achieving better workload balance and greater fulfillment responsiveness. Additionally, due-date flexibility has a substantial impact on scheduling performance, emphasizing its critical role in maintaining customer satisfaction and delivery reliability. Overall, this study presents a novel modeling perspective and an intelligent optimization framework for outbound scheduling and resource coordination in Q-commerce, providing both theoretical and practical insights for developing responsive and sustainable instant retail logistics systems.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"75 ","pages":"Article 101571"},"PeriodicalIF":6.3,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Forgiveness catalyst or risk amplifier? The impact of AI customer service anthropomorphism on consumers’ forgiveness willingness 宽恕催化剂还是风险放大器？人工智能客服拟人化对消费者宽恕意愿的影响

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2026-01-01 DOI: 10.1016/j.elerap.2026.101572

Huawei Liu , Xiqing Han , Sihong Li , Xiyang Zhao , Min Zhang

In the business domain, AI customer service has become increasingly prevalent, yet service failures remain inevitable. A key challenge lies in mitigating the negative emotions that arise from such failures. Given that anthropomorphism significantly shapes consumer–AI interactions, it is essential to examine its influence on service failure outcomes and consumers’ willingness to forgive. This study investigates the degree of anthropomorphism in AI customer service as the independent variable, consumers’ forgiveness willingness toward companies as the dependent variable, and aversion to AI service failures as the mediating variable. Additionally, task criticality and relational norms are introduced as moderating variables. A research model is developed and tested through one pilot and three formal scenario-based experiments, followed by data analysis using SPSS. The results reveal four key findings: (1) In service failure contexts, a higher degree of anthropomorphism in AI customer service positively influences consumers’ willingness to forgive. (2) Aversion to AI mediates the relationship between anthropomorphism and forgiveness; specifically, higher anthropomorphism leads to lower aversion and, consequently, greater forgiveness. (3) Task criticality moderates this effect: under high task criticality, anthropomorphic AI increases forgiveness even in the face of failure, whereas under low task criticality, anthropomorphism has no significant effect on forgiveness. (4) Relational norms also moderate the effect. In public relationships (e.g., long-term or communal exchanges), higher anthropomorphism enhances forgiveness. However, in exchange-based relationships, anthropomorphism does not alleviate negative emotions or promote forgiveness. These findings offer theoretical insights and practical implications for the anthropomorphic design of AI systems and for managing customer relationships in AI-mediated service contexts.

在商业领域，人工智能客户服务变得越来越普遍，但服务失败仍然不可避免。一个关键的挑战在于减轻这些失败带来的负面情绪。鉴于拟人化极大地塑造了消费者与人工智能的互动，有必要研究其对服务失败结果和消费者原谅意愿的影响。本研究以人工智能客户服务中的拟人化程度为自变量，消费者对企业的宽恕意愿为因变量，对人工智能服务失败的厌恶为中介变量。此外，还引入了任务关键性和关系规范作为调节变量。通过一个试点和三个正式的基于场景的实验，开发和测试了一个研究模型，然后使用SPSS进行数据分析。研究结果揭示了四个主要发现：(1)在服务失败情境下，人工智能客户服务中更高程度的拟人化对消费者的原谅意愿有积极影响。(2)对人工智能的厌恶在拟人化与宽恕的关系中起中介作用；具体来说，更高的拟人化导致更低的厌恶，因此，更大的宽恕。(3)任务临界性调节了这一效应：在高任务临界性条件下，拟人化人工智能即使面对失败也能提高宽恕，而在低任务临界性条件下，拟人化人工智能对宽恕没有显著影响。(4)关系规范也有调节作用。在公共关系中（例如，长期或公共交流），更高的拟人化提高了宽恕。然而，在以交换为基础的关系中，拟人化并不能缓解负面情绪或促进宽恕。这些发现为人工智能系统的拟人化设计以及在人工智能介导的服务环境中管理客户关系提供了理论见解和实践意义。

{"title":"Forgiveness catalyst or risk amplifier? The impact of AI customer service anthropomorphism on consumers’ forgiveness willingness","authors":"Huawei Liu , Xiqing Han , Sihong Li , Xiyang Zhao , Min Zhang","doi":"10.1016/j.elerap.2026.101572","DOIUrl":"10.1016/j.elerap.2026.101572","url":null,"abstract":"<div><div>In the business domain, AI customer service has become increasingly prevalent, yet service failures remain inevitable. A key challenge lies in mitigating the negative emotions that arise from such failures. Given that anthropomorphism significantly shapes consumer–AI interactions, it is essential to examine its influence on service failure outcomes and consumers’ willingness to forgive. This study investigates the degree of anthropomorphism in AI customer service as the independent variable, consumers’ forgiveness willingness toward companies as the dependent variable, and aversion to AI service failures as the mediating variable. Additionally, task criticality and relational norms are introduced as moderating variables. A research model is developed and tested through one pilot and three formal scenario-based experiments, followed by data analysis using SPSS. The results reveal four key findings: (1) In service failure contexts, a higher degree of anthropomorphism in AI customer service positively influences consumers’ willingness to forgive. (2) Aversion to AI mediates the relationship between anthropomorphism and forgiveness; specifically, higher anthropomorphism leads to lower aversion and, consequently, greater forgiveness. (3) Task criticality moderates this effect: under high task criticality, anthropomorphic AI increases forgiveness even in the face of failure, whereas under low task criticality, anthropomorphism has no significant effect on forgiveness. (4) Relational norms also moderate the effect. In public relationships (e.g., long-term or communal exchanges), higher anthropomorphism enhances forgiveness. However, in exchange-based relationships, anthropomorphism does not alleviate negative emotions or promote forgiveness. These findings offer theoretical insights and practical implications for the anthropomorphic design of AI systems and for managing customer relationships in AI-mediated service contexts.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"75 ","pages":"Article 101572"},"PeriodicalIF":6.3,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Integrating customer preferences into operational decision-making for prioritizing emerging technologies in last-mile delivery 将客户偏好整合到运营决策中，以优先考虑最后一英里交付中的新兴技术

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2026-01-01 DOI: 10.1016/j.elerap.2026.101573

Ahmet Çalık , Esra Boz , Sinan Çizmecioğlu , Erfan Babaee Tirkolaee

The rapid expansion of e-commerce has positioned last-mile delivery as the most critical and resource-intensive stage of modern supply chains. Firms must balance multiple and often conflicting objectives: reducing costs, minimizing environmental impacts, and meeting growing customer expectations for speed, reliability, and personalization. While previous research has focused on operational efficiency and routing optimization, limited attention has been given to frameworks that integrate customer preferences with technology-enabled decision-making. This study develops a hybrid decision support model by extending the classical Simple Weight Calculation (SIWEC) method with grey numbers (G-SIWEC), capable of handling uncertainty in subjective judgments to generate robust criterion weights. These weights are incorporated into a multi-objective optimization model, solved using the Weighted Sum Scalarization Method (WSSM), to minimize delivery costs and emissions while maximizing service quality. The model explicitly considers emerging delivery technologies, including drone, Autonomous Vehicle (AV), and bicycle (e-bike), to explore innovative, sustainable, and customer-centric delivery strategies. Findings highlight technology-specific patterns: drone and bicycle excel in lightweight, eco-friendly deliveries, AV dominates mid-range logistics, and conventional truck remains indispensable for heavy loads. By linking customer preferences to technology-driven operational decisions, this work provides practical insights for managers and policymakers seeking to design efficient, sustainable, and innovative last-mile delivery systems for e-commerce. These implications should be interpreted within the context of the numerical experiment conducted in this study.

电子商务的快速发展使最后一英里配送成为现代供应链中最关键、资源最密集的阶段。公司必须平衡多个经常相互冲突的目标：降低成本，最小化环境影响，满足客户对速度、可靠性和个性化日益增长的期望。虽然之前的研究主要集中在运营效率和路线优化上，但对将客户偏好与技术支持决策相结合的框架的关注有限。本研究通过扩展经典的灰色数简单权重计算（G-SIWEC）方法，开发了一种混合决策支持模型，能够处理主观判断中的不确定性，生成鲁棒的准则权重。这些权重被纳入到一个多目标优化模型中，使用加权和标量化方法（WSSM）求解，以最小化交付成本和排放，同时最大化服务质量。该模型明确考虑了新兴的配送技术，包括无人机、自动驾驶汽车（AV）和自行车（e-bike），以探索创新、可持续和以客户为中心的配送策略。研究结果强调了技术特定模式：无人机和自行车在轻型环保配送方面表现出色，无人驾驶汽车主导中程物流，而传统卡车在重载运输方面仍然不可或缺。通过将客户偏好与技术驱动的运营决策联系起来，这项工作为寻求为电子商务设计高效、可持续和创新的最后一英里交付系统的管理人员和政策制定者提供了实用的见解。这些含义应该在本研究中进行的数值实验的背景下解释。

{"title":"Integrating customer preferences into operational decision-making for prioritizing emerging technologies in last-mile delivery","authors":"Ahmet Çalık , Esra Boz , Sinan Çizmecioğlu , Erfan Babaee Tirkolaee","doi":"10.1016/j.elerap.2026.101573","DOIUrl":"10.1016/j.elerap.2026.101573","url":null,"abstract":"<div><div>The rapid expansion of e-commerce has positioned last-mile delivery as the most critical and resource-intensive stage of modern supply chains. Firms must balance multiple and often conflicting objectives: reducing costs, minimizing environmental impacts, and meeting growing customer expectations for speed, reliability, and personalization. While previous research has focused on operational efficiency and routing optimization, limited attention has been given to frameworks that integrate customer preferences with technology-enabled decision-making. This study develops a hybrid decision support model by extending the classical Simple Weight Calculation (SIWEC) method with grey numbers (G-SIWEC), capable of handling uncertainty in subjective judgments to generate robust criterion weights. These weights are incorporated into a multi-objective optimization model, solved using the Weighted Sum Scalarization Method (WSSM), to minimize delivery costs and emissions while maximizing service quality. The model explicitly considers emerging delivery technologies, including drone, Autonomous Vehicle (AV), and bicycle (e-bike), to explore innovative, sustainable, and customer-centric delivery strategies. Findings highlight technology-specific patterns: drone and bicycle excel in lightweight, eco-friendly deliveries, AV dominates mid-range logistics, and conventional truck remains indispensable for heavy loads. By linking customer preferences to technology-driven operational decisions, this work provides practical insights for managers and policymakers seeking to design efficient, sustainable, and innovative last-mile delivery systems for e-commerce. These implications should be interpreted within the context of the numerical experiment conducted in this study.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"75 ","pages":"Article 101573"},"PeriodicalIF":6.3,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977834","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A scalable framework for ranking integration in large-scale online reviews: Integrating clustering and multi-attribute decision-making 大规模在线评论排名集成的可扩展框架：集成聚类和多属性决策

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-12-27 DOI: 10.1016/j.elerap.2025.101570

Mengchun Ma , Bin Yu , Zeshui Xu

The proliferation of online reviews has transformed consumer decision-making in the hospitality industry, yet information overload remains a critical challenge. This study proposes a novel hotel ranking framework that integrates cluster analysis with Multi-Attribute Decision-Making (MADM) methods to address the issue of preference differences in large-scale user rating data. By leveraging the X-means algorithm, large-scale user evaluations are partitioned into clusters with homogeneous preferences, enabling efficient intra-cluster ranking via the TOPSIS method. A hybrid weighting mechanism, combining group size and information entropy, ensures balanced aggregation of rankings across clusters. The proposed dominance-non-dominance degree metric synthesizes global preferences, offering a robust solution for ranking 16 five-star hotels in London. Comparative analyses with traditional MADM methods demonstrate superior consistency (Kendall’s

τ = 0.983

, Spearman’s

ρ = 0.997

) and stability across diverse datasets. This framework not only streamlines decision-making for travelers but also provides actionable insights for hoteliers to refine service quality and market positioning.

在线评论的激增改变了酒店业消费者的决策，但信息过载仍然是一个严峻的挑战。本研究提出了一种新的酒店排名框架，将聚类分析与多属性决策（MADM）方法相结合，以解决大规模用户评级数据中的偏好差异问题。通过利用x均值算法，大规模用户评价被划分为具有同质偏好的聚类，通过TOPSIS方法实现高效的聚类内排名。混合加权机制，结合组大小和信息熵，确保跨集群的排名均衡聚集。所提出的主导-非主导度度量综合了全球偏好，为伦敦16家五星级酒店的排名提供了一个强大的解决方案。与传统的MADM方法比较分析表明，在不同的数据集上，Kendall 's τ=0.983, Spearman 's ρ=0.997)具有更好的一致性和稳定性。这个框架不仅简化了旅行者的决策，还为酒店经营者提供了可操作的见解，以改善服务质量和市场定位。

{"title":"A scalable framework for ranking integration in large-scale online reviews: Integrating clustering and multi-attribute decision-making","authors":"Mengchun Ma , Bin Yu , Zeshui Xu","doi":"10.1016/j.elerap.2025.101570","DOIUrl":"10.1016/j.elerap.2025.101570","url":null,"abstract":"<div><div>The proliferation of online reviews has transformed consumer decision-making in the hospitality industry, yet information overload remains a critical challenge. This study proposes a novel hotel ranking framework that integrates cluster analysis with Multi-Attribute Decision-Making (MADM) methods to address the issue of preference differences in large-scale user rating data. By leveraging the X-means algorithm, large-scale user evaluations are partitioned into clusters with homogeneous preferences, enabling efficient intra-cluster ranking via the TOPSIS method. A hybrid weighting mechanism, combining group size and information entropy, ensures balanced aggregation of rankings across clusters. The proposed dominance-non-dominance degree metric synthesizes global preferences, offering a robust solution for ranking 16 five-star hotels in London. Comparative analyses with traditional MADM methods demonstrate superior consistency (Kendall’s <span><math><mrow><mi>τ</mi><mo>=</mo><mn>0</mn><mo>.</mo><mn>983</mn></mrow></math></span>, Spearman’s <span><math><mrow><mi>ρ</mi><mo>=</mo><mn>0</mn><mo>.</mo><mn>997</mn></mrow></math></span>) and stability across diverse datasets. This framework not only streamlines decision-making for travelers but also provides actionable insights for hoteliers to refine service quality and market positioning.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"75 ","pages":"Article 101570"},"PeriodicalIF":6.3,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841472","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Direct or agency? Manufacturer encroachment under platform’s data-driven marketing 直接还是代理？平台数据驱动营销下的厂商蚕食

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-12-05 DOI: 10.1016/j.elerap.2025.101567

Renji Duan , Zhenzhong Guan , Xinlan Ye , Jianbiao Ren

The development of e-commerce has enabled manufacturers to wholesale products to retail platforms, which then resell them to consumers. However, manufacturers may enter the end market by either establishing an independent official retail store (direct encroachment) or selling directly to consumers on the platform (agency encroachment). In response to the competitive pressure caused by such encroachment, the platform is motivated to invest in data-driven marketing (DDM) to enhance consumer purchasing utility. To this end, we analyse the interaction between the manufacturer’s encroachment strategy and the platform’s DDM decision using a game-theoretic model. The key results are as follows. First, the manufacturer prefers direct encroachment when the commission rate is high and selling cost is low, and agency encroachment when the commission rate is low. The decision depends on two opposing effects: the competition and expansion effects. Second, DDM is not always effective in deterring encroachment. Under certain conditions, DDM may instead induce the manufacturer to introduce a direct channel. Third, as the commission rate increases, the equilibrium outcome may evolve from “DDM + no encroachment” to “no DDM + agency encroachment”, and then back to “DDM + no encroachment”. Interestingly, during these transitions, the manufacturer’s (platform’s) profit increases (decreases) abruptly. Finally, we further explore six extensions and validate the robustness of our main conclusions.

电子商务的发展使制造商能够将产品批发给零售平台，然后再转售给消费者。然而，制造商进入终端市场的方式可能是建立独立的官方零售店（直接侵占），也可能是在平台上直接向消费者销售（代理侵占）。为了应对这种侵蚀带来的竞争压力，平台有动机投资数据驱动营销（DDM），以提高消费者的购买效用。为此，我们利用博弈论模型分析了制造商的入侵策略与平台的DDM决策之间的相互作用。主要结果如下。首先，在佣金率高、销售成本低的情况下，制造商倾向于直接侵占，而在佣金率低的情况下，制造商倾向于代理侵占。这一决定取决于两种相反的效应：竞争效应和扩张效应。其次，DDM在阻止入侵方面并不总是有效。在某些条件下，DDM可能会促使制造商引入直接渠道。第三，随着佣金率的增加，均衡结果可能会从“DDM +无侵犯”演变为“无DDM +代理侵犯”，再回到“DDM +无侵犯”。有趣的是，在这些转变过程中，制造商（平台）的利润会突然增加（减少）。最后，我们进一步探讨了六个扩展，并验证了我们主要结论的鲁棒性。

{"title":"Direct or agency? Manufacturer encroachment under platform’s data-driven marketing","authors":"Renji Duan , Zhenzhong Guan , Xinlan Ye , Jianbiao Ren","doi":"10.1016/j.elerap.2025.101567","DOIUrl":"10.1016/j.elerap.2025.101567","url":null,"abstract":"<div><div>The development of e-commerce has enabled manufacturers to wholesale products to retail platforms, which then resell them to consumers. However, manufacturers may enter the end market by either establishing an independent official retail store (direct encroachment) or selling directly to consumers on the platform (agency encroachment). In response to the competitive pressure caused by such encroachment, the platform is motivated to invest in data-driven marketing (DDM) to enhance consumer purchasing utility. To this end, we analyse the interaction between the manufacturer’s encroachment strategy and the platform’s DDM decision using a game-theoretic model. The key results are as follows. First, the manufacturer prefers direct encroachment when the commission rate is high and selling cost is low, and agency encroachment when the commission rate is low. The decision depends on two opposing effects: the competition and expansion effects. Second, DDM is not always effective in deterring encroachment. Under certain conditions, DDM may instead induce the manufacturer to introduce a direct channel. Third, as the commission rate increases, the equilibrium outcome may evolve from “DDM + no encroachment” to “no DDM + agency encroachment”, and then back to “DDM + no encroachment”. Interestingly, during these transitions, the manufacturer’s (platform’s) profit increases (decreases) abruptly. Finally, we further explore six extensions and validate the robustness of our main conclusions.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"75 ","pages":"Article 101567"},"PeriodicalIF":6.3,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145705615","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

From clicks to context: A heterogeneous graph framework for diagnosing consumer shopping goals and personalizing retail strategy 从点击到上下文：用于诊断消费者购物目标和个性化零售策略的异构图形框架

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-11-01 DOI: 10.1016/j.elerap.2025.101557

Yongjie Yan , Hui Xie

In modern e-commerce, recommender systems are vital for personalization. However, many systems exhibit “contextual blindness,” failing to distinguish between fundamental user motivations like established brand affinity and exploratory category-seeking. This limitation leads to suboptimal recommendations and missed revenue opportunities. To address this gap, we propose the Heterogeneous Graph Context-Aware Recommender (HGCAR). The framework constructs a multi-relational graph of users, items, brands, and categories. It employs a hierarchical attention mechanism to not only predict user choices but also to diagnose the underlying drivers by quantifying the influence of each context (e.g., brand vs. category) for each user. The resulting user-specific attention weights (

β

) function as managerially interpretable diagnostics. This allows practitioners to segment users based on their primary purchasing drivers (e.g., “Brand Loyalists” vs. “Category Explorers”), enabling the deployment of highly targeted marketing campaigns. The proposed framework is evaluated on large-scale Amazon datasets. Results show that HGCAR achieves significant improvements in recommendation accuracy over state-of-the-art baselines. Furthermore, an illustrative simulation suggests that segmenting users with our diagnostic weights has the potential for substantial increases in marketing campaign Return on Investment (ROI). This work bridges the gap between predictive accuracy and managerial actionability, transforming recommendation engines from black-box predictors into strategic decision tools for personalized marketing and inventory optimization.

在现代电子商务中，推荐系统是实现个性化的关键。然而，许多系统表现出“上下文盲目性”，无法区分基本的用户动机，如建立品牌亲和力和探索性类别寻找。这种限制导致了次优推荐和错失收入机会。为了解决这一差距，我们提出了异构图上下文感知推荐器（HGCAR）。该框架构建了用户、项目、品牌和类别的多关系图。它采用了一种分层注意机制，不仅可以预测用户的选择，还可以通过量化每个用户的每个上下文（例如，品牌与类别）的影响来诊断潜在的驱动因素。由此产生的用户特定关注权重（β）作为管理上可解释的诊断。这允许从业者根据用户的主要购买驱动因素（例如，“品牌忠诚者”与“品类探索者”）对用户进行细分，从而实现高度针对性的营销活动的部署。在大规模的Amazon数据集上对该框架进行了评估。结果表明，与最先进的基线相比，HGCAR在推荐精度方面取得了显着提高。此外，一个说明性模拟表明，用我们的诊断权重对用户进行细分，有可能大幅提高营销活动的投资回报率（ROI）。这项工作弥合了预测准确性和管理可操作性之间的差距，将推荐引擎从黑箱预测器转变为个性化营销和库存优化的战略决策工具。

{"title":"From clicks to context: A heterogeneous graph framework for diagnosing consumer shopping goals and personalizing retail strategy","authors":"Yongjie Yan , Hui Xie","doi":"10.1016/j.elerap.2025.101557","DOIUrl":"10.1016/j.elerap.2025.101557","url":null,"abstract":"<div><div>In modern e-commerce, recommender systems are vital for personalization. However, many systems exhibit “contextual blindness,” failing to distinguish between fundamental user motivations like established brand affinity and exploratory category-seeking. This limitation leads to suboptimal recommendations and missed revenue opportunities. To address this gap, we propose the Heterogeneous Graph Context-Aware Recommender (HGCAR). The framework constructs a multi-relational graph of users, items, brands, and categories. It employs a hierarchical attention mechanism to not only predict user choices but also to diagnose the underlying drivers by quantifying the influence of each context (e.g., brand vs. category) for each user. The resulting user-specific attention weights (<span><math><mi>β</mi></math></span>) function as managerially interpretable diagnostics. This allows practitioners to segment users based on their primary purchasing drivers (e.g., “Brand Loyalists” vs. “Category Explorers”), enabling the deployment of highly targeted marketing campaigns. The proposed framework is evaluated on large-scale Amazon datasets. Results show that HGCAR achieves significant improvements in recommendation accuracy over state-of-the-art baselines. Furthermore, an illustrative simulation suggests that segmenting users with our diagnostic weights has the potential for substantial increases in marketing campaign Return on Investment (ROI). This work bridges the gap between predictive accuracy and managerial actionability, transforming recommendation engines from black-box predictors into strategic decision tools for personalized marketing and inventory optimization.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"74 ","pages":"Article 101557"},"PeriodicalIF":6.3,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Consumer purchase intention during brand crisis: A study on intangible cultural heritage brands in the context of e-commerce live streaming 品牌危机下的消费者购买意愿：电商直播背景下的非遗品牌研究

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-11-01 DOI: 10.1016/j.elerap.2025.101560

Wei Zhang, Hao Ran, Yuan Gong, Xiaohui Zhou

Live-stream commerce—marked by real-time interactivity, immersive visuals, and embedded social cues—has become mainstream. While it expands exposure, its speed and reach can also amplify negative sentiment, posing acute risks for intangible cultural heritage (ICH) brands that bear commercial and cultural responsibilities. Using the Stimulus–Organism–Response framework, we examine how live-shopping experience and brand-crisis perception shape purchase intention via parallel mediators: brand trust (cognitive) and ICH-preservation awareness (affective). Focusing on Pien Tze Huang—an ICH-listed, time-honoured Chinese brand—we triangulate literature review, field investigation, and a survey (N = 432). Structural-equation modelling with multiple mediation yields three findings. (i) Engaging live-shopping elevates professional trust and deepens cultural identification, jointly increasing purchase intention. (ii) Crisis perception erodes trust yet heightens awareness of cultural scarcity; under certain conditions the latter dominates, paradoxically raising purchase intention. (iii) Trust and ICH-preservation awareness transmit these influences in parallel, underscoring intertwined cognition and emotion. Managerially, ICH brands should pair interactive, evidence-first livestreams with transparent, multi-platform communication; explain pricing while demonstrating craft and provenance; and ensure compliant, traceable supply chains that reconcile commercial logic with heritage stewardship. These conclusions are analytical generalisations from a high-salience ICH setting; portability depends on heritage depth and verifiable transparency. Future research should adopt longitudinal or cross-case designs across ICH categories and cultural contexts, incorporating behavioural traces to test boundary conditions and strengthen external validity.

以实时交互性、身临其境的视觉效果和嵌入的社交线索为标志的直播商业已经成为主流。在扩大曝光率的同时，它的速度和影响范围也会放大负面情绪，给承担商业和文化责任的非物质文化遗产（ICH）品牌带来严重风险。采用刺激-机体-反应框架，我们考察了现场购物体验和品牌危机感知如何通过平行中介：品牌信任（认知）和文化遗产保护意识（情感）来塑造购买意愿。我们以片仔癀这一在香港上市、历史悠久的中国品牌为研究对象，采用文献综述、实地调查和问卷调查三种方法（N = 432）。具有多重中介的结构方程建模产生了三个发现。(1)参与现场购物提升了专业信任，加深了文化认同，共同提高了购买意愿。（ii）危机感侵蚀了信任，但又增强了文化稀缺意识；在一定条件下，后者占主导地位，矛盾地提高了购买意愿。（iii）信任和文化遗产保护意识并行传递这些影响，强调认知和情感相互交织。在管理上，ICH品牌应该将互动的、证据优先的直播与透明的、多平台的传播相结合；在展示工艺和产地的同时解释价格；并确保兼容的、可追溯的供应链，使商业逻辑与遗产管理相协调。这些结论是来自一个高度突出的非物质文化遗产环境的分析性概括；可移植性取决于遗产的深度和可验证的透明性。未来的研究应采用跨非物质文化遗产类别和文化背景的纵向或跨案例设计，结合行为痕迹来测试边界条件并加强外部有效性。

{"title":"Consumer purchase intention during brand crisis: A study on intangible cultural heritage brands in the context of e-commerce live streaming","authors":"Wei Zhang, Hao Ran, Yuan Gong, Xiaohui Zhou","doi":"10.1016/j.elerap.2025.101560","DOIUrl":"10.1016/j.elerap.2025.101560","url":null,"abstract":"<div><div>Live-stream commerce—marked by real-time interactivity, immersive visuals, and embedded social cues—has become mainstream. While it expands exposure, its speed and reach can also amplify negative sentiment, posing acute risks for intangible cultural heritage (ICH) brands that bear commercial and cultural responsibilities. Using the Stimulus–Organism–Response framework, we examine how live-shopping experience and brand-crisis perception shape purchase intention via parallel mediators: brand trust (cognitive) and ICH-preservation awareness (affective). Focusing on Pien Tze Huang—an ICH-listed, time-honoured Chinese brand—we triangulate literature review, field investigation, and a survey (N = 432). Structural-equation modelling with multiple mediation yields three findings. (i) Engaging live-shopping elevates professional trust and deepens cultural identification, jointly increasing purchase intention. (ii) Crisis perception erodes trust yet heightens awareness of cultural scarcity; under certain conditions the latter dominates, paradoxically raising purchase intention. (iii) Trust and ICH-preservation awareness transmit these influences in parallel, underscoring intertwined cognition and emotion. Managerially, ICH brands should pair interactive, evidence-first livestreams with transparent, multi-platform communication; explain pricing while demonstrating craft and provenance; and ensure compliant, traceable supply chains that reconcile commercial logic with heritage stewardship. These conclusions are analytical generalisations from a high-salience ICH setting; portability depends on heritage depth and verifiable transparency. Future research should adopt longitudinal or cross-case designs across ICH categories and cultural contexts, incorporating behavioural traces to test boundary conditions and strengthen external validity.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"74 ","pages":"Article 101560"},"PeriodicalIF":6.3,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145571235","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Understanding the knowledge sharing behaviors in social Commerce: Affordances, coactive vicarious Learning, and need for cognitive closure 理解社交商务中的知识共享行为：支持、协同替代学习和认知封闭需求

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-11-01 DOI: 10.1016/j.elerap.2025.101565

Zhiyuan Nong, Jing Wu

This research investigates how social media affordances (SMA) influence social commerce customers’ knowledge sharing through their engagement in coactive vicarious learning (CVL). We extend the theory of SMA by incorporating the privacy-controlling ability afforded by social media. Five complementary forms of SMA then drive CVL, with knowledge sharing as the learning outcome. Specifically, this study examines the dual-stage moderating role of the need for cognitive closure (NCC), revealing how the extended SMA affects users with different psychological states. A quantitative survey was conducted on several leading Chinese social commerce platforms to test the research model. Data from 867 respondents were analyzed using the PLS-SEM method. The results confirm that: (1) SMA has a positive effect on CVL, and CVL positively affects knowledge sharing; (2) CVL partially mediates the relationship between the extended SMA and knowledge sharing; (3) NCC acts as a dual-stage positive moderator in the indirect effect of SMA on knowledge sharing through CVL. This research contributes to the literature on knowledge sharing and social commerce, deepens the understanding of consumer learning, and advances the application of cognitive psychology in the study of digital learning and consumer behavior.

本研究探讨了社交媒体支持（SMA）如何通过参与协同替代学习（CVL）来影响社交商务客户的知识共享。我们通过纳入社交媒体提供的隐私控制能力来扩展SMA理论。然后，五种互补形式的SMA驱动CVL，以知识共享作为学习结果。具体而言，本研究考察了认知闭合需求（NCC）的双阶段调节作用，揭示了扩展SMA如何影响不同心理状态的用户。我们对中国几家领先的社交商务平台进行了定量调查，以检验研究模型。使用PLS-SEM方法分析了867名受访者的数据。结果表明：(1)SMA对知识共享有正向影响，知识共享对知识共享有正向影响；(2) CVL在扩展SMA与知识共享的关系中起部分中介作用；(3) NCC在SMA对CVL知识共享的间接影响中起双阶段正向调节作用。本研究扩充了关于知识共享和社交商务的文献，加深了对消费者学习的理解，推进了认知心理学在数字学习和消费者行为研究中的应用。

{"title":"Understanding the knowledge sharing behaviors in social Commerce: Affordances, coactive vicarious Learning, and need for cognitive closure","authors":"Zhiyuan Nong, Jing Wu","doi":"10.1016/j.elerap.2025.101565","DOIUrl":"10.1016/j.elerap.2025.101565","url":null,"abstract":"<div><div>This research investigates how social media affordances (SMA) influence social commerce customers’ knowledge sharing through their engagement in coactive vicarious learning (CVL). We extend the theory of SMA by incorporating the privacy-controlling ability afforded by social media. Five complementary forms of SMA then drive CVL, with knowledge sharing as the learning outcome. Specifically, this study examines the dual-stage moderating role of the need for cognitive closure (NCC), revealing how the extended SMA affects users with different psychological states. A quantitative survey was conducted on several leading Chinese social commerce platforms to test the research model. Data from 867 respondents were analyzed using the PLS-SEM method. The results confirm that: (1) SMA has a positive effect on CVL, and CVL positively affects knowledge sharing; (2) CVL partially mediates the relationship between the extended SMA and knowledge sharing; (3) NCC acts as a dual-stage positive moderator in the indirect effect of SMA on knowledge sharing through CVL. This research contributes to the literature on knowledge sharing and social commerce, deepens the understanding of consumer learning, and advances the application of cognitive psychology in the study of digital learning and consumer behavior.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"74 ","pages":"Article 101565"},"PeriodicalIF":6.3,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145684779","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Privacy protection investment decisions of e-commerce platforms under mean-variance preferences 均值方差偏好下电子商务平台的隐私保护投资决策

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-11-01 DOI: 10.1016/j.elerap.2025.101563

JianHua Wang, Bin Zhao

As data privacy issues become increasingly prominent in the digital economy, differences in privacy protection investment and risk preferences among supply chain members frequently lead to decision-making conflicts. This study considers a two-tier supply chain consisting of a supplier and an e-commerce platform, where the platform may unintentionally cause consumer privacy breaches and faces a certain probability of being held accountable and compensating consumers. The objective is to explore the decision-making mechanisms of supply chain members under different risk preferences and privacy protection investment modes, and to analyze the impact of privacy protection investment and risk preferences on supply chain members by considering the compensation cost associated with privacy breaches. We derive the following main conclusions: regardless of whether the platform invests in privacy protection, an increase in the probability of privacy breaches

f_{1}

suppresses market demand, lowers the wholesale price, and reduces the profits of both the supplier and the platform. Given a certain level of protection capability and technological efficiency, the e-commerce platform should actively invest in privacy protection. The platform’s risk attitude affects its pricing strategy and profit distribution structure; although a risk-averse platform tends to set a lower price to expand market demand, its subjective utility is generally lower than that of a risk-neutral platform.

随着数字经济中数据隐私问题的日益突出，供应链成员在隐私保护投入和风险偏好方面的差异经常导致决策冲突。本研究考虑一个由供应商和电子商务平台组成的双层供应链，其中平台可能无意中造成消费者隐私泄露，并面临一定的责任追究和赔偿消费者的可能性。目的是探索不同风险偏好和隐私保护投资模式下供应链成员的决策机制，并通过考虑与隐私泄露相关的补偿成本来分析隐私保护投资和风险偏好对供应链成员的影响。我们得出以下主要结论：无论平台是否投入隐私保护，隐私泄露概率的增加f1都会抑制市场需求，降低批发价格，降低供应商和平台的利润。在具备一定保护能力和技术效率的情况下，电子商务平台应积极投入隐私保护。平台的风险态度影响其定价策略和利润分配结构；虽然风险厌恶型平台倾向于设定较低的价格来扩大市场需求，但其主观效用普遍低于风险中性平台。

{"title":"Privacy protection investment decisions of e-commerce platforms under mean-variance preferences","authors":"JianHua Wang, Bin Zhao","doi":"10.1016/j.elerap.2025.101563","DOIUrl":"10.1016/j.elerap.2025.101563","url":null,"abstract":"<div><div>As data privacy issues become increasingly prominent in the digital economy, differences in privacy protection investment and risk preferences among supply chain members frequently lead to decision-making conflicts. This study considers a two-tier supply chain consisting of a supplier and an e-commerce platform, where the platform may unintentionally cause consumer privacy breaches and faces a certain probability of being held accountable and compensating consumers. The objective is to explore the decision-making mechanisms of supply chain members under different risk preferences and privacy protection investment modes, and to analyze the impact of privacy protection investment and risk preferences on supply chain members by considering the compensation cost associated with privacy breaches. We derive the following main conclusions: regardless of whether the platform invests in privacy protection, an increase in the probability of privacy breaches <span><math><msub><mi>f</mi><mn>1</mn></msub></math></span> suppresses market demand, lowers the wholesale price, and reduces the profits of both the supplier and the platform. Given a certain level of protection capability and technological efficiency, the e-commerce platform should actively invest in privacy protection. The platform’s risk attitude affects its pricing strategy and profit distribution structure; although a risk-averse platform tends to set a lower price to expand market demand, its subjective utility is generally lower than that of a risk-neutral platform.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"74 ","pages":"Article 101563"},"PeriodicalIF":6.3,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145684777","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Mitigating domain adaptation and cold start challenges in cross-domain recommender systems using generative adversarial network model 利用生成对抗网络模型缓解跨域推荐系统中的域适应和冷启动挑战

IF 6.3 3区管理学 Q1 BUSINESS

Electronic Commerce Research and Applications

Pub Date : 2025-11-01 DOI: 10.1016/j.elerap.2025.101564

Matthew O. Ayemowa , Roliana Ibrahim , Noor Hidayah Zakaria

Recommender systems often face challenges of cold start, specifically when expanding into new domains. The current research has shown the successful impact of Generative Adversarial Networks (GANs) on domain adaptation problems. However, the underutilization of GAN-based model needs more attention for more personalized recommendations. Domain adaptation helps to mitigate these issues by transferring knowledge from a source domain to a target domain. This paper proposes a novel approach that leverages Generative Adversarial Networks (GANs) to enhance domain adaptation in Cross-domain recommender systems (CDRS). The proposed model, Domain Adaptation Cross-domain Generative Adversarial Networks (DAC-GAN) with the incorporation of auxiliary information utilized the generator to produce synthetic user-item interactions in the target domain and a discriminator to distinguish between real and generated interactions, thereby improving the performance of the recommendation. By integrating auxiliary information into GANs model, the framework bridges the domain gap, and this enables accurate predictions in the target domain. Comprehensive experiments on benchmark datasets: Amazon, Movielens, Yelp and Book-crossing demonstrate the effectiveness of the proposed technique, by achieving significant improvements in the evaluation metrics: Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) compared to existing techniques. Experiments conducted on benchmark datasets demonstrate that DAC-GAN outperforms the existing methods in terms of recommendation accuracy and adaptability.

推荐系统经常面临冷启动的挑战，特别是在扩展到新领域时。目前的研究表明，生成对抗网络（GANs）在领域自适应问题上取得了成功。然而，对于更个性化的推荐，gan模型的利用不足值得关注。领域适应通过将知识从源领域转移到目标领域来帮助缓解这些问题。本文提出了一种利用生成对抗网络（GANs）增强跨域推荐系统（CDRS）的域适应性的新方法。所提出的领域自适应跨领域生成对抗网络（DAC-GAN）模型结合了辅助信息，利用生成器在目标领域产生合成的用户-物品交互，并利用鉴别器区分真实交互和生成交互，从而提高了推荐的性能。通过将辅助信息集成到gan模型中，该框架弥补了领域差距，从而实现了目标领域的准确预测。在Amazon、Movielens、Yelp和Book-crossing等基准数据集上进行的综合实验证明了该技术的有效性，与现有技术相比，该技术在评估指标：均方根误差（RMSE）和平均绝对误差（MAE）方面取得了显著改进。在基准数据集上进行的实验表明，DAC-GAN在推荐准确率和自适应性方面优于现有方法。

{"title":"Mitigating domain adaptation and cold start challenges in cross-domain recommender systems using generative adversarial network model","authors":"Matthew O. Ayemowa , Roliana Ibrahim , Noor Hidayah Zakaria","doi":"10.1016/j.elerap.2025.101564","DOIUrl":"10.1016/j.elerap.2025.101564","url":null,"abstract":"<div><div>Recommender systems often face challenges of cold start, specifically when expanding into new domains. The current research has shown the successful impact of Generative Adversarial Networks (GANs) on domain adaptation problems. However, the underutilization of GAN-based model needs more attention for more personalized recommendations. Domain adaptation helps to mitigate these issues by transferring knowledge from a source domain to a target domain. This paper proposes a novel approach that leverages Generative Adversarial Networks (GANs) to enhance domain adaptation in Cross-domain recommender systems (CDRS). The proposed model, Domain Adaptation Cross-domain Generative Adversarial Networks (DAC-GAN) with the incorporation of auxiliary information utilized the generator to produce synthetic user-item interactions in the target domain and a discriminator to distinguish between real and generated interactions, thereby improving the performance of the recommendation. By integrating auxiliary information into GANs model, the framework bridges the domain gap, and this enables accurate predictions in the target domain. Comprehensive experiments on benchmark datasets: Amazon, Movielens, Yelp and Book-crossing demonstrate the effectiveness of the proposed technique, by achieving significant improvements in the evaluation metrics: Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) compared to existing techniques. Experiments conducted on benchmark datasets demonstrate that DAC-GAN outperforms the existing methods in terms of recommendation accuracy and adaptability.</div></div>","PeriodicalId":50541,"journal":{"name":"Electronic Commerce Research and Applications","volume":"74 ","pages":"Article 101564"},"PeriodicalIF":6.3,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145618095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0