1. Introduction
1. 引言
With the ongoing advancement of next-generation information technologies, such as big data, cloud computing, and artificial intelligence (AI), the robotics industry is increasingly integrating sophisticated intelligent capabilities. AI robots based on intelligent services are gradually becoming a key focus for businesses. artificial intelligence (AI) robots refer to intelligent devices capable of interacting with users, communicating, and providing services through autonomous and highly adaptive interfaces powered by AI systems[1]. In 2019, the Shenzhen Intercontinental Hotel, in partnership with Shenzhen Telecom and Huawei, launched the world's first 5G smart hotel. The hotel equipped its lobby with AI robots to assist guests with services such as information retrieval and wayfinding, enhancing both interactive experience and service quality. Today, numerous hotel chains worldwide, including Huazhu, Atour, Marriott, Hilton, and InterContinental, are utilizing AI robots to provide services like the delivery of goods and food[2]. These applications demonstrate the potential of AI robots to reduce labor and operational costs and promote service efficiency[3], making them highly valued across various service industries.
随着大数据、云计算和人工智能 （AI） 等下一代信息技术的不断进步，机器人行业越来越多地集成复杂的智能功能。 基于智能服务的 AI 机器人正逐渐成为企业关注的重点 。 人工智能 （AI） 机器人是指能够通过由 AI 系统提供支持的自主和高度自适应的界面与用户交互、通信和提供服务的智能设备 [1].2019 年，深圳洲际酒店与深圳电信和华为合作，推出了全球首家 5G 智慧酒店。酒店大堂配备了人工智能机器人 ，以协助客人进行信息检索和寻路等服务，从而提升互动体验和服务质量。如今，全球众多连锁酒店 ，包括华住、爱途、万豪、希尔顿和洲际 ， 都在使用 AI 机器人提供商品和食品 配送等服务 [2]。 这些应用展示了 AI 机器人在降低劳动力和运营成本以及提高服务效率方面的潜力 [3]，使其在各种服务行业中受到高度重视。
However, given the inherent heterogeneity and personalization of services[4], combined with the current limitations of AI technology, AI robots may misunderstand customer intent or provide mismatched responses, leading to service failures[5]. Such failures can impair corporate profitability[6] and customer satisfaction[7]. In light of the growing number of service failures in recent years, it has become imperative to explore efficient remedial measures AI robots can conduct to restore consumer trust.
然而，鉴于服务固有的异质性和个性化 [4]，再加上当前 AI 技术的局限性，AI 机器人可能会误解客户的意图或提供不匹配的响应，从而导致服务失败 [5]。 此类失败会损害公司盈利能力 [6] 和客户满意度 [7]。 鉴于近年来服务故障的数量不断增加 ， 探索人工智能机器人可以采取有效的补救措施以恢复消费者的信任已成为一项重要的事情。
Currently there are two main approaches regarding service recovery: human staff recovery and independent recovery by AI robots.. Some scholars suggested that AI robots can apologize for service failures. However, subsequent investigation has shown that compared to human employees, apologies delivered by robots may be perceived as lacking in sincerity[8]. Other research indicated that AI robots can mitigate the impact of service failures through empathetic responses to some extent[9]. Empathy can enhance consumer trust in AI robots and increase their willingness to continue using them. Furthermore, when AI robots display empathy, consumers may unconsciously suppress negative emotions and generate positive feelings, such as satisfaction and happiness[3]. According to social exchange theory, the positive emotions elicited in consumers by empathetic interactions may foster greater tolerance toward both the robot and the company, leading to a reciprocal display of goodwill. However, most existing studies on empathic robots center around customer retention following service failures[8,10], with little attention given to how empathy influences consumer forgiveness.
目前关于服务恢复有两种主要方法： 人工人员重新协调和由 AI 机器人独立恢复 。。一些学者建议 AI 机器人可以为服务失败道歉 。 然而， 随后的调查 表明， 与人类员工相比，机器人的道歉可能被认为缺乏诚意 [8]。 其他研究表明 ，AI 机器人可以在 一定程度上通过移情反应来减轻服务故障的影响 [9]。 同理心可以增强消费者对 AI 机器人的信任，并增加他们继续使用它们的意愿。此外，当 AI 机器人表现出同理心时，消费者可能会无意识地抑制负面情绪并产生积极的感觉 ， 例如满足感和幸福感 [3]。 根据社会交换理论 ， 消费者通过同理心互动引发的积极情绪可能会培养对机器人和公司更大的宽容，从而导致互惠的善意展示 。 然而，大多数关于移情机器人的现有研究都集中在服务失败后的客户保留上[8,10]，而很少关注同理心如何影响消费者的宽恕。
Thus, drawing on social exchange theory, this study examined the impact of empathetic responses from AI robots on consumers from two key aspects: the positive emotions and trust elicited. Through three experiments, the study investigated the mechanisms underlying this effect and assessed the effectiveness of empathetic responses from AI robots under varying levels of service failures. The findings contribute to refining the research framework for AI service applications, advancing service marketing theory in sectors such as hospitality, dining, and tourism, and providing valuable insights into businesses seeking to optimize the design of AI robots.
因此， 基于社会交换理论 ， 本研究从 两个关键方面考察了人工智能机器人的移情反应 对消费者的影响 ：引发的积极情绪和信任 。通过三个实验，该研究调查了这种效应背后的机制，并评估了 AI 机器人在不同级别的服务失败下移情反应的有效性 。这些发现有助于完善 AI 服务应用的研究框架，推进酒店、餐饮和旅游等领域的服务营销理论，并为 寻求优化 AI 机器人设计的企业提供有价值的见解 。
2. Literature Review
2. 文献综述
2.1 Research on AI Robots in the Service Industry
2.1 服务业 AIRobot 研究
The early stages of research focused on classifying the application of AI robots in the service industry from various perspectives. An extensively adopted classification framework is the service recipient and task type matrix developed by Lovelock & C.H. (1983), which categorizes service robots into two main types: physical service robots and virtual service robots[11]. Huang and Rust (2018) identified four types of AI based on service nature and the types of intelligence required: mechanical, analytical, intuitive, and emotional. This typology has become prevalent in recent literature on AI applications in the service industry[12].
早期研究的重点是从不同角度对 AI 机器人在服务行业的应用进行分类。 广泛采用的分类框架是由 Lovelock 和 C.H.（1983）开发的服务接收者和任务类型矩阵，它将服务机器人分为两种主要类型：物理服务机器人和虚拟服务机器人 [11]。 Huang 和 Rust （2018） 根据服务性质和所需的智能类型确定了四种类型的 AI：机械、分析、直观和情感。这种类型在最近关于服务行业 AI 应用的文献中变得普遍 [12]。
As AI technology continues to be integrated into service operations, studies are increasingly addressing the adoption of AI by both consumers and businesses. From the perspective of consumers, much of the research has centered on evaluating consumer acceptance, willingness to use, and satisfaction with AI robots. A variety of conceptual models have been developed to identify factors that influence consumers' willingness to adopt new technologies, such as perceived usefulness and ease of use[13], social influence[14], cognitive processes[15], experiences[16], and hedonic motivations[17].
随着 AI 技术不断集成到服务运营中，越来越多的研究涉及消费者和企业对 AI 的采用 。 从 消费者的角度来看 ，大部分研究都集中在评估消费者对 AI 机器人的接受度、使用意愿和满意度 上。已经开发了各种概念模型来识别影响消费者采用新技术意愿的因素，例如感知有用性和易用性 [13]、社会影响 [14]、认知过程 [15]、体验 [16] 和享乐动机 [17]。
With the deepening of the understanding of AI robot services, scholars have recognized that consumers expect AI robots to deliver services comparable to those of human employees. This has led to a growing body of literature on the anthropomorphic design of AI robots, examining factors such as appearance[18], voice and language style[19], head tilts, and smiles[20] in an effort to advance consumer experience and satisfaction.
随着 对 AI 机器人服务理解的加深 ，学者们认识到，消费者期望 AI 机器人能够提供与人类员工相当的服务。这导致越来越多的关于 AI 机器人拟人化设计的文献， 研究了外表 [18]、声音和语言风格 [19]、头部倾斜和微笑 [20] 等因素， 以努力提高消费者体验和满意度。
However, while anthropomorphism can improve consumer perceptions of AI robots, more researchers have noted that emotional responses from AI robots during service interactions do not always meet consumer expectations. In particular, AI robots often struggle with flexible emotional interaction and are perceived to lack human warmth, which limits their effectiveness in service encounters.
然而，虽然拟人化可以改善消费者对 AI 机器人的看法 ，但研究人员指出，AI 机器人在服务交互过程中的情绪反应并不总是符合消费者的期望。特别是，AI 机器人经常难以进行灵活的情感互动，并且被认为缺乏人类的热情，这限制了它们在服务遭遇中的有效性。
In summary, existing studies on AI in services primarily involved robot classification, consumer acceptance, satisfaction evaluations, and initial usage intentions. However, there is limited research on how consumers psychologically respond to AI robots' remedial actions in the context of service failures and whether they would forgive a service failure based on the robot's performance. This study aimed to address this gap by exploring effective remedial strategies for AI robots in service failure situations, thus mitigating consumer dissatisfaction and obtaining their forgiveness.
综上所述，现有的 人工智能在服务中的研究主要涉及机器人分类、消费者接受度、满意度评价和初始使用意向。然而，关于消费者在服务故障的情况下如何对 AI 机器人的补救措施做出心理反应 ，以及他们是否会根据机器人的表现原谅服务故障，研究有限。 本研究旨在通过探索 AI 机器人在服务故障情况下 的有效补救策略来解决这一差距 ， 从而减轻消费者的不满并获得他们的原谅。
2.2 Service Recovery
2.2 服务恢复
Service recovery refers to the actions taken by service providers to address service failures, resolve consumer complaints, and mitigate negative consequences[21]. Effective service recovery can help businesses obtain consumer forgiveness and repair their relationship with customers[22]. Existing literature on human service recovery has identified two primary types of recovery strategies: psychological symbolic recovery (including empathy, apologies, and promises) and material compensation recovery, which involves monetary or tangible compensations (such as refunds or discounts)[23-25]
服务恢复是指服务提供商为解决服务故障 、解决消费者投诉和减轻负面后果而采取的行动 [21]。 有效的服务恢复可以帮助企业获得消费者的宽恕，并重新建立与客户的 IR 关系 [22]。 关于人类服务恢复的大量文献确定了两种主要类型的恢复策略：心理象征性恢复 （ 包括同理心、道歉和承诺 ） 和物质补偿恢复，涉及金钱或有形补偿 （ 如退款或折扣 ）[2325].
However, due to the modular nature of current AI service systems and the heterogeneity of customers in the service industry, AI systems face challenges in accurately assessing customers' psychological expectations. Traditional human recovery methods may not be applicable to the field of AI services. Therefore, it is essential to develop strategies tailored to the distinct technology characteristics of AI robots.
然而，由于当前 AI 服务系统的模块化特性以及服务行业客户的异质性 ，AI 系统在准确评估客户的心理期望方面面临挑战。 传统的人类恢复方法可能不适用于 AI 服务领域 。因此，必须根据 AI 机器人的独特技术特性制定量身定制的策略 。
A new line of thought has emerged from Lv et al., who explored the role of AI emotional intelligence in service recovery. They found that AI empathy, as an aspect of emotional intelligence, can significantly enhance consumers' willingness to continue using AI services and improve the effectiveness of service recovery efforts[9]. This insight opens up possibilities for exploring how AI robots can independently manage service recovery. This study furthered this discussion by focusing on its impact on customer trust and emotions and exploring how AI empathy influences consumer forgiveness.
Lv 等人提出了一种新的思路 。，他探讨了 AI 情商在服务恢复中的作用。他们发现，AI 同理心作为情商的一个方面，可以显著提高消费者继续使用 AI 服务的意愿，并提高服务恢复工作的有效性 [9]。 这一洞察为探索 AI 机器人如何独立管理服务恢复提供了可能性。本研究通过关注其对客户信任和情感的影响 ，并探索 AI 同理心如何影响消费者的宽恕 ，进一步推动了这一讨论。
2.3 Social Exchange Theory
Social exchange theory is a fundamental framework for understanding human behavior in social interactions. It posits that individuals engage in exchanges to fulfill their needs, aiming to maximize their benefits[26]. The decision to engage in such exchanges is based on the perceived balance of benefits and costs. Perceived benefits can be divided into extrinsic benefits (e.g., money, services, and products) and internal rewards (e.g., psychological satisfaction and a sense of recognition). In social interactions, individuals maintain relationships by reciprocating the value they have received. This reciprocity fosters positive, mutually beneficial interactions. The need for reciprocity is what drives social exchange behavior[27].
社会交换理论是理解人类在社会互动中的行为的基本框架。它假设个人参与交换以满足他们的需求，旨在最大化他们的利益 [26]。 参与此类交换的决定是基于对收益和成本的感知平衡 。 感知的好处可以分为外在利益 （例如，金钱、服务和 产品）和内部奖励（例如，心理满意度和认可感）。 在社交互动中，个人通过回报他们所获得的价值来维持关系。这种互惠互利促进了积极、互利的互动。对互惠的需求是驱动社会交换行为的原因 [27]。
According to social exchange theory, the success of AI robots in offering service recovery hinges on whether the recovery efforts provide sufficient perceived benefits to the consumers. Trust in AI is defined as, when confronted with risk, consumers' confidence in AI to fulfill its commitments and safeguard their interests, reflecting their expectations of the extrinsic benefits that AI can offer[28]. Additionally, when AI robots deliver empathetic responses, consumers are more likely to diminish their negative emotions and experience intrinsic benefits, such as satisfaction and happiness[3]. These benefits can profoundly influence consumer attitudes and behaviors.
与社会交换理论相呼应，AI 机器人在提供服务恢复方面的成功取决于恢复工作是否为消费者提供了足够的感知利益 。对 AI 的信任 被定义为 ，当面临风险时 ， 消费者对 AI 履行承诺和维护其利益的信心 ， 反映了他们对 AI 可以提供的外在好处的期望 [28]。 此外，当 AI 机器人提供移情反应时 ，消费者更有可能减少他们的负面情绪并体验到内在的好处 ，例如满意度和幸福感 [3]。 这些好处可以深刻影响消费者的态度和行为。
Thus, based on the social exchange theory, this paper believes that, following service failures, AI robots' empathetic responses for service recovery can affect consumer trust and emotions, which in turn elicits their forgiveness behavior.
因此，基于 社会交换理论 ，本文认为 ，在服务失败后 ， 人工智能机器人对服务恢复的移情反应 会影响消费者的信任和情绪，进而引发他们的宽恕行为。
3. Research Hypothesis
3. 研究假设
3.1 The Impact of AI Empathetic Responses on Consumer Forgiveness
3.1 AI 情感 Responses 对消费者宽恕的影响
Empathy represents the emotional response that arises from an individual's understanding of another person's emotional state, where the response mirrors or aligns with the other person's actual or potential feelings[29]. Empathy plays a critical role not only in improving interpersonal relationships but also in resolving social conflicts and fostering cultural understanding. AI empathy refers to the ability of AI systems to recognize and respond to human emotions and behaviors, signifying the social capabilities of AI systems[30]. Unlike human empathy, which encompasses a wide range of emotional expressions, such as languages, words, touches, and comforting gestures, AI empathy is constrained by technological limitations and the costs associated with integrating such features into service systems. However, empathetic responses, as a remedial action, can effectively avoid the issues mentioned above. They can be carried out through various interactions with consumers and have become the main application of AI empathy[31].
同理心代表了个人对他人情绪状态的理解而产生的情绪反应 ，其中反应反映或与其他人的实际或潜在感受保持一致 [29]。 同理心不仅在改善人际关系方面发挥着关键作用，而且在解决社会冲突和促进文化理解方面也起着关键作用。AI 同理心是指 AI 系统识别和响应人类情绪和行为的能力， 表示 AI 系统的社交能力 [30]。 与人类同理心不同，人类同理心包含广泛的情感表达 ， 例如语言 、 文字、 触摸和安慰手势 ， 而 AI 同理心则受到技术限制以及与将此类功能集成到服务系统中的相关成本的限制。 然而， 同理心反应作为一种补救措施，可以有效避免上述问题。 它们可以通过与消费者的各种互动来实现 ，并已成为 AI 同理心的主要应用 [31]。
Upon service failures, companies' primary focus is on mitigating the negative emotions of customers. During interactions with humans, AI empathy signals that the AI understands consumers' perspectives and prioritizes their interests, which helps alleviate their unpleasant feelings[32].
在服务失败时 ，公司的主要重点是 减轻 客户的负面情绪 。 在与人类互动的过程中 ，AI 同理心表明 AI 理解消费者的观点并优先考虑他们的兴趣，这有助于减轻他们的不愉快感觉 [32]。
According to social exchange theory, individuals will reciprocate the value they have received to maintain mutually beneficial relationships[33]. Consequently, this article hypothesizes that after a service failure, consumers who receive high-empathetic responses from AI robots will experience more positive perceptions and emotions, which will, in turn, foster higher forgiveness toward the AI robots and the company as a form of "reciprocal returns."
根据社会交换理论 ，个人将回报他们所获得的价值以维持互惠互利的关系 [33]。 因此 ， 本文假设在服务失败后，从 AI 机器人那里获得高度同理心反应的消费者将体验到更积极的感知和情绪，这反过来又会培养对 AI 机器人的高度宽恕 s 和公司作为 “ 互惠回报 s.“
Therefore, the following hypothesis is proposed:
因此，提出了以下假设 ：
H1: Under the service failures of AI robots, high-empathetic responses will lead to greater consumer forgiveness compared to low-empathetic responses.
H1： 在 AI 机器人的服务失败下 ，与低同理心反应相比，高同理心反应会导致更大的消费者宽恕 。
3.2 The Mediating Role of Trust and Emotions
3.2 信任和情绪的中介作用
Trust is a pivotal element in human behavior and social interaction. In marketing field, it has been demonstrated that trust is usually built through the transactional relationship involved in the exchange of goods. Moreover, trust plays a crucial role in mitigating the negative impact of service failures and sustaining robust business-consumer relationships[34-36]. It has been reported that when AI robots exhibit mechanical, repetitive error messages or provide low-empathetic responses, it may lead to customer dissatisfaction and undermine their trust[37]. In contrast, high-empathetic responses can alleviate the sense of detachment often associated with AI robots, providing feedback that mirrors the warmth and understanding typical of human interactions. Lv et al. confirmed that AI robots with high-empathetic responses can substantially strengthen customer trust, which in turn enhances their willingness to continue using the service[38]. This can reinforce consumers' positive perceptions of AI robots' ability to safeguard consumer interests, thereby elevating their trust. Elevated trust, in turn, fosters a more forgiving attitude toward AI robots and their service failures.
信任是 人类行为和社会互动的关键要素。在营销领域 ， 已经 证明信任通常是 通过商品交换所涉及的交易关系建立的。此外， 信任在减轻服务故障的负面影响和维持强大的业务与消费者关系方面发挥着至关重要的作用 [34， 36]。 据报道 ，当 AI 机器人表现出机械、重复的错误信息或提供低同理心的响应时 ， 可能会导致客户的不满并破坏他们的信任 [37]。 相比之下，高度同理心的反应可以缓解通常与 AI 机器人相关的疏离感 ， 提供反映人类互动中典型的温暖和理解的反馈 。 Lv 等人。 证实具有高同理心反应的 AI 机器人可以大大增强客户的信任，进而增强他们继续使用该服务的意愿 [38]。 这可以加强消费者对 AI 机器人维护消费者利益 能力的积极认知，从而提高他们的信任度。反过来，增强的信任会培养对 AI 机器人及其服务失败的更宽容的态度 。
Based on this, this paper hypothesizes the following:
基于此， 本文假设如下：
H2a: In the context of AI service failures, consumers will exhibit higher trust in AI robots that offer high-empathetic responses than those that provide low-empathetic responses.
H2a：在 AI 服务失败的情况下，消费者对提供高移情反应的 AI 机器人的信任度会 高于提供低移情反应的 AI 机器人 。
H2b: Trust mediates the relationship between AI empathetic responses and consumer forgiveness.
H2b：信任在 AI 移情反应和消费者宽恕之间的关系之间起中介作用。
Emotion is defined as "a psychological state that stems from the cognitive evaluation of an event or thought, potentially leading to specific behaviors"[39]. Research has verified that AI systems capable of resonating with users, demonstrating supportive behaviors, and perceiving emotions can enhance users' positive emotions[40]. When robots recognize and respond appropriately to others' emotions, they are considered to have empathy, which fosters a greater sense of belonging in users. In contrast, a lack of empathy can yield negative outcomes, such as misunderstandings and hostility[41]. Westbrook identified the relationships between six post-purchase emotions and three behaviors: positive emotions are positively correlated with satisfaction ratings and favorable word-of-mouth, and they are negatively associated with customer complaints. In contrast, negative emotions are positively correlated with complaints and negative word-of-mouth, and they are negatively linked to satisfaction[42]. According to social exchange theory, high-empathetic responses from AI robots can trigger positive emotions in consumers, thus encouraging consumers to reciprocate through forgiveness.
情绪被定义为“一种心理状态， 它源于对事件或思想的认知评估， 可能导致 特定行为”[39]。 研究证实 ，能够与用户产生共鸣、展示支持行为和感知情绪的 AI 系统可以增强用户的积极情绪 [40]。 当机器人识别并适当地回应他人的情绪时， 它们被认为具有同理心，从而在用户中培养更大的归属感 。相反，缺乏同理心会产生 负面结果，例如误解和敌意 [41]。Westbrook 确定了 6 种售后情绪与 3 种行为之间的关系：积极情绪与满意度和良好口碑呈正相关 ， 与顾客投诉呈负相关 。相比之下， 负面情绪与抱怨和负面口碑呈正相关 ， 与满意度呈负相关 [42]。 根据社会交换理论 ，AI 机器人的高移情反应可以触发消费者的积极情绪 ， 从而鼓励消费者通过宽恕来回报。
Based on this, this paper posits the following hypotheses:
基于此，这是论文 posits 以下假设：
H3a: In the context of AI service failures, high-empathetic replies are more likely to evoke positive emotions in consumers than low-empathetic ones.
H3a：在 AI 服务失败的背景下 ， 高同理心的回复比低同理心的回复更有可能在消费者中唤起积极的情绪 。
H3b: Emotions mediate the effect of empathetic responses from AI robots on consumer forgiveness.
H3b：情绪介导 AI 机器人的移情反应对消费者宽恕的影响。
3.3 The Moderating Role of the Severity of Service Failures
3.3 S 服务严重程度的调节作用 Failures
The severity of service failures denotes the extent to which consumers perceive a service failure[43]. The greater the perceived loss, the higher the severity of the failure. Numerous studies have shown that the severity of service failures plays a decisive role in the effectiveness of recovery efforts[19,44,45]. Additionally, the severity of service failures significantly influences both the evaluation of the service provider and consumer satisfaction[46-47]. In the hotel industry, prior research indicated that higher severity of service failures is typically correlated with lower customer satisfaction[43]. Moreover, the severity of service failures can markedly impact consumer emotions. More significant failures tend to evoke stronger negative emotions in consumers, such as disappointment and anger[48-51]. As the severity of service failures increases, consumers' tolerance for deviations from expectations narrows, further exacerbating dissatisfaction[52].
服务故障的严重性表示消费者感知服务故障的程度 [43]。 感知到的损失越大，故障的严重性就越高。大量研究表明， 服务故障的严重性对恢复工作的有效性起着决定性的作用 [19,44,45]。 此外， 服务故障的严重性会显著影响对服务提供商的评估和消费者满意度 [46-47]。 在酒店行业， 先前的研究表明 ， 服务故障的严重性越高 ， 客户满意度 就越低 [43]。 此外， 服务故障的严重性会显著影响消费者的情绪。 许多重大失败往往会在消费者中引起更强烈的负面情绪，例如失望和愤怒 [48-51]。 随着服务故障的严重性增加，消费者对偏离期望的容忍度缩小，进一步加剧了不满 [52]。
Given these insights, this article hypothesizes that when the severity of service failures is high, empathetic responses from AI robots may be insufficient to meet consumers' recovery expectations, thus failing to alleviate their negative emotions.
鉴于这些见解， 本文假设当服务故障的严重程度 很高时，AI 机器人的移情反应可能足以满足消费者的康复期望， 从而无法缓解他们的负面情绪。
Therefore, this paper proposes the following hypothesis:
因此， 本文提出了以下假设：
H4: The severity of service failures moderates the impact of empathetic responses from AI robots on trust (H4a), emotions (H4b), and consumer forgiveness (H4c).
H4： 服务故障的严重性缓和了 AI 机器人的移情反应对信任 （H4a）、情绪 （H4b） 和消费者宽恕 （H4c） 的影响。
Specifically, with low service failure severity, high-empathetic responses will induce higher trust, more positive emotions, and greater consumer forgiveness than low-empathetic responses. Conversely, when the severity of service failures is high, empathetic responses have a slight effect on consumer forgiveness, and the mediating role of trust and emotions becomes insignificant.
具体来说， 在服务故障严重性较低的情况下， 与低同理心反应相比 ， 高同理心反应会诱导更高的信任、更积极的情绪和更大的消费者宽恕 。 相反 ， 当服务故障的严重程度 较高时 ， 移情反应 s 对消费者的宽恕性影响较小 ，信任和情感的中介作用变得显著。
Based on the hypotheses outlined above and social exchange theory, the following theoretical framework is established, as depicted in Figure 1:
基于上述假设和社会交换理论，建立了以下理论框架，如图 1 所示：
4. Research Design and Data Analysis
4. 研究设计和数据分析
4.1 Experiment 1
4.1 实验 1
4.1.1 Experimental Design
4.1.1 实验设计
Experiment 1 adopted a two-level, between-subjects design (empathetic responses: high vs. low), investigating the influence of empathetic responses from AI robots on consumer forgiveness following service failures.
实验 1 采用了两级、受试者间设计 （ 移情反应 s： 高与低）， 研究了 AI 机器人的 移情反应对服务失败后消费者宽恕的影响 。
4.1.2 Pre-Test
4.1.2 Pre-T est
The pre-test aimed to validate the manipulation effectiveness of the independent variable, empathetic responses. Before the formal experiment, the pre-test was conducted on the experimental materials of the empathetic responses from AI robots. The two experimental conditions only differ in the content of the responses, while all other aspects remain identical. The scenario was adapted from the study by Lü Xingyang et al.[9], simulating a service failure in a hotel with an AI robot.
预测试旨在验证自变量移情反应 s 的纵有效性 。在正式实验之前， 对 AI 机器人的移情反应的实验材料进行了 预测试 。这两个实验条件仅在响应的内容上有所不同 ，而所有其他方面都相同。该情景改编自 Lü Xingyang 等人的研究。[9]，使用 AI 机器人模拟酒店的服务故障 。
In this scenario:
在此方案中：
The customer checked into a hotel with an intelligent service system while traveling. He could not find the tissues and slippers. He instructed the in-room AI: "I need a pack of tissues and a pair of slippers." The doorbell rang, and the AI robot delivered the requested items. However, he discovered that the robot only brought him the tissues and not the slippers. At this point, the robot interface prompted him to evaluate the service, and he selected the option "This task was not successfully completed."
客户在旅行中入住了一家拥有智能服务系统的酒店。 他找不到纸巾和拖鞋。 他指示房间里的 AI：“我需要一包纸巾和一双拖鞋。门铃响了 ，AI 机器人交付了请求的物品。然而， 他发现机器人只给他带来了纸巾，没有给他带来拖鞋。此时，机器人界面提示他评估服务， 他选择了 “ 此任务未成功完成” 选项 。“
Regarding this, a low-empathetic response from the robot was: "Dear customer, we apologize for not completing the service successfully." A high-empathetic response was: "Darling, you're the cutest person I've ever met! I know I'm a bit clumsy, but please, lovely you, give me a chance to make it right!"
对此， 机器人的回应非常感同身受 地回答道：“ 尊敬的客户，对于未能成功完成服务，我们深表歉意。 一个非常同情的回答是：“ 亲爱的，你是我见过的最可爱的人！我知道我有点笨拙，但拜托，可爱的你，给我一个机会来纠正它！
The pre-test was conducted using Wenjuanxing, an online survey platform, with 135 participants recruited (54.1% female, 84.4% aged 19-29). Participants were randomly assigned to one of the two experimental conditions (high vs. low empathy). After reading the scenario, they were asked to rate the empathy level of the AI robot's response.
预测试是使用在线调查平台文娟星进行的，招募了 135 名参与者（54.1% 为女性，84.4% 为 19-29 岁）。参与者被随机分配到两个实验条件之一（高同理心与低同理心）。阅读场景后，他们被要求对 AI 机器人反应的同理心水平进行评分 。
The four measurement items were adapted from Ronan et al.[53]:
这四个测量项目 改编自 Ronan 等人。[53]：
"I feel that the AI robot considered my feelings in its response."
“我觉得 AI 机器人在回应时考虑了我的感受。”
"I feel that the AI robot centered its response around me."
“我觉得 AI 机器人的反应以我为中心。”
"I feel that the AI robot understood my perspective."
“我觉得 AI 机器人理解我的观点。”
"I feel that the AI robot took my emotions into account in its response."
“我觉得 AI 机器人在回应时考虑到了我的情绪。”
These items were scored on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree).
这些项目采用 7 分李克特量表评分（1 = strongly disagree，7 = strongly agree）。
The results of the independent samples t-test revealed that the high-empathetic response group rated the robot's response significantly higher than the low-empathetic response group (M_low empathy = 2.96, SD = 1.073 vs. M_high empathy = 5.45, SD = 0.085; t(115.325) = 16.043, p < 0.001). The materials met the requirements for the experimental manipulation.
独立样本 t 检验的结果显示 ，高移情反应组对机器人反应的评分明显高于低移情反应组（M_low 同理心 = 2.96，SD = 1.073 vs. M_high 同理心 = 5.45，SD = 0.085;t（115.325） = 16.043，p < 0.001）。这些材料满足实验作的要求 。
4.1.3 Formal Experiment
4.1.3 形式实验
The formal experiment closely followed the methodology used in the pre-test. Conducted on the Wenjuanxing online survey platform, the study recruited 223 participants (56.1% female, 62.3% aged 19-29) for paid participation. Participants were randomly assigned to one of two experimental conditions (high empathy vs. low empathy) and asked to read the experimental scenario and complete a questionnaire. The questionnaire included measures of service failure severity, empathy level, consumer forgiveness, and demographic information (gender and age).
正式实验严格遵循预测试中使用的方法。该研究在文娟星在线调查平台上进行，招募了 223 名参与者（56.1% 为女性，62.3% 为 19-29 岁）进行付费参与。参与者被随机分配到两个实验条件之一（高同理心与低同理心），并被要求阅读实验场景并完成问卷。问卷包括服务失败严重程度、同理心水平、消费者宽恕度和人口统计信息（性别和年龄）的测量。
The empathy measurement used in this experiment was identical to the pre-test. Consumer forgiveness was assessed based on scales from Muhammad, Gul-E-Rana[54], and Suri et al.[55], with modifications tailored to the experimental scenario. The forgiveness scale consisted of six items:
本实验中使用的同理心测量与预测试相同。消费者宽恕是根据 Muhammad、Gul-E-Rana[54] 和 Suri 等人的量表评估的。[55]，并根据实验场景进行了修改。宽恕量表由六个项目组成：
"I will not take my frustration out on the service provider because of this service failure."
“我不会因为这次服务失败而将我的挫败感发泄在服务提供商身上。”
"I will not develop a dislike for the service provider because of this service failure."
“我不会因为这次服务失败而不喜欢服务提供商。”
"I am letting go of my negative feelings."
“我正在放下我的负面情绪。”
"I will forgive this service failure."
“我会原谅这次服务失败。”
"I will give the robot an opportunity to make amends."
“我会给机器人一个补偿的机会。”
"I am willing to continue using services from this robot."
“我愿意继续使用这个机器人的服务。”
Each item was rated on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree).
每个项目都按照 7 分李克特量表进行评分（1 = strongly disagree，7 = strongly agree）。
Service failure severity was evaluated with a single item: "How severe do you perceive this service failure to be?" (7-point Likert scale, 1 = not severe at all, 7 = very severe).
使用单个项目评估服务故障严重性 ：“您认为此服务故障有多严重？（7 点李克特量表，1 = 根本不严重，7 = 非常严重）。
4.1.4 Data Analysis
4.1.4 数据分析
Manipulation Check
作检查
The results of the independent samples t-test indicated a significant difference in the perceived empathy between the two experimental groups. The high-empathetic response group rated the AI's empathy remarkably higher than the low-empathetic response group (M_low empathy = 3.156, SD = 1.114 vs. M_high empathy = 5.491, SD = 1.098; t(221) = 15.742, p < 0.001).
独立样本 t 检验的结果表明 ，两个实验组之间的感知同理心存在显著差异。高同理心反应组对 AI 的同理心评分明显高于低同理心反应组（M_low 同理心 = 3.156，SD = 1.114 vs. M_high 同理心 = 5.491，SD = 1.098;t（221） = 15.742，p < 0.001）。
Regarding service failure severity, no significant difference was found between the two groups (M_low empathy = 3.83, SD = 1.599 vs. M_high empathy = 3.80, SD = 1.489; t(221) = 0.131, p = 0.896). The potential impact of service failure severity on the experiment was controlled. The situational stimuli were effective, and the experimental manipulation was effective.
关于服务故障的严重程度，两组之间没有发现显著差异 （M_low 同理心 = 3.83，SD = 1.599 vs. M_high 同理心 = 3.80，SD = 1.489;t（221） = 0.131，p = 0.896）。 控制了服务故障严重性对实验的潜在影响 。情境刺激 有效 ， 实验作有效。
Main Effect Test
主效测试
With empathetic responses as the independent variable and consumer forgiveness as the dependent variable, the independent samples t-test was performed, comparing forgiveness willingness between the two groups with varying levels of empathy. The results revealed that, upon service failures, the high-empathetic response group demonstrated significantly greater forgiveness intentions than the-low-empathetic response group (M_low empathy = 3.922, SD = 1.320 vs. M_high empathy = 5.673, SD = 0.925; t(208.178) = 11.556, p < 0.001). This provides preliminary support for H1.
以 empathetic responses 为自变量，以消费者宽恕为因变量 ， 进行自样本 t 检验 ， 比较不同同理心水平两组的宽恕意愿 。结果显示， 在服务失败时 ，高移情反应组表现出明显高于低移情反应组的宽恕意向 （M_low 同理心 = 3.922，SD = 1.320 vs. M_high 同理心 = 5.673，SD = 0.925;t（208.178） = 11.556，p < 0.001）。这为 H1 提供了初步支持。
4.1.5 Discussion
4.1.5 讨论
The findings from Experiment 1 suggest that AI empathetic responses play a significant role in influencing consumer forgiveness following service failures. Specifically, higher empathetic responses from AI robots lead to greater forgiveness from consumers compared to lower empathetic responses. H1 is validated.
实验 1 的结果表明，AI 移情反应在影响服务失败后的消费者宽恕方面发挥着重要作用 。具体来说， 与较低的移情反应 s 相比，AI 机器人的较高移情反应 s 导致消费者的宽恕度更高 。
4.2 Experiment 2
4.2 实验 2
4.2.1 Experiment Design
4.2.1 实验设计
Experiment 2 employed a single-factor, two-level (empathetic responses: high vs. low) between-subjects design. The objective was to examine the impact of AI robots' empathetic responses on consumer forgiveness after service failures and the mediating role of trust and emotions.
实验 2 采用单因素、两级 （ 移情反应 s：高与低） 受试者间设计。目的是研究 AI 机器人的移情反应对服务失败后消费者宽恕的影响 ，以及信任和情感的中介作用 。
4.2.2 Formal Experiment
4.2.2 正式实验
The experimental scenario was adapted from Lü Xingyang et al.[9] to simulate a service failure in a hotel environment.
实验场景改编自 Lü Xingyang 等人。[9] 来模拟酒店环境中的服务故障。
The scenario is as follows:
方案如下所示：
The customer was staying at a hotel with an intelligent service system during his trip. After a full rest in his room, he decided to go to the hotel restaurant but was unsure of the location. He approached an AI robot in the lobby and asked: "Where can I eat in the hotel?" The robot responded with a lengthy description: "This hotel offers a Chinese restaurant, a Western restaurant, a dessert buffet area, 200 individual seats, and 240 group seats. The facilities included are as follows..." He quickly interrupted and inquired again: "Where is the restaurant?"
该客户在旅途中入住了一家拥有智能服务系统的酒店 。 在房间里休息了一整天 后， 他决定去酒店的餐厅，但不确定位置。 他走近大堂里的 ed 一个 AI 机器人 ，问 ed：“ 我在酒店哪里可以吃饭？“ 机器人以冗长的描述回应道：” 这家酒店提供中餐厅、西餐厅、甜点自助餐区、200 个独立座位和 240 个团体座位 。 包括的设施如下 ...“ 他赶紧打断 ed， 又问道：' 餐厅在哪里？“
In this case, a low-empathetic response from the robot: "Sorry, I didn't quite catch that. Could you please repeat your question?" A high-empathetic response from the robot: "Sorry, I didn't hear that properly. I understand you might be getting a bit impatient, but please give me another chance to assist you."
在这种情况下，机器人的低同理心回应 ：“ 对不起，我没有完全明白。您能否重复您的问题？“ 机器人高度同理心的回应 ：” 对不起，我没有听清楚。我知道您可能有点不耐烦，但请给我另一个机会来帮助您。“
He reiterated: "Which floor is the restaurant on?"
他重申：“餐厅在哪一层？
Regarding this, a low-empathetic response from the robot: "Dear guest, the restaurant is located on the second floor. Enjoy your meal." A high-empathetic response from the robot: "Dear guest, your delicious meal is waiting for you on the second floor. I hope my earlier confusion didn't spoil your appetite!"
对此， 机器人以善解人意的回答道 ：“ 亲爱的客人，餐厅位于二楼。请享用。“ 机器人高度同理心的回应 ：” 亲爱的客人，您的美味佳肴在二楼等着您。希望我刚才的困惑没有破坏你的胃口！“
This experiment was conducted through Wenjuanxing (a Chinese survey platform), with 206 participants (49% female, 82.5% aged 19-29) who were recruited for paid participation. Participants were randomly assigned to one of the two experimental groups (high vs. low empathy) and asked to read the experimental scenario and complete a questionnaire. The survey included measures for service failure severity, empathy level, trust, emotion, consumer forgiveness, and demographic information (gender and age).
该实验是通过文娟星（一个中国调查平台）进行的，招募了 206 名参与者（49% 为女性，82.5% 为 19-29 岁）参与。参与者被随机分配到两个实验组之一（高同理心与低同理心），并被要求阅读实验场景并完成问卷。该调查包括服务失败的严重程度、同理心水平、信任、情感、消费者宽恕和人口统计信息（性别和年龄）的衡量标准。
The measurement items for service failure severity, empathy level, and consumer forgiveness willingness were consistent with those used in Experiment 1. The trust items were adjusted according to scales developed by Flavian et al.[56], Siguaw et al.[57], and Park and Tribe[58], consisting of five statements (7-point Likert scale, 1 = Strongly Disagree, 7 = Strongly Agree):
服务失败严重程度、同理心水平和消费者宽恕意愿的测量项目与实验 1 中使用的项目一致。根据 Flavian 等人开发的量表对信任项目进行调整 。[56]，Siguaw 等人。[57]， 以及 Park and Tribe[58]，由五个陈述组成（7 点李克特量表，1 = 非常不同意，7 = 非常同意）：
"I believe this robot has the ability to resolve issues encountered during service."
“我相信这个机器人有能力解决服务过程中遇到的问题 。“
"I believe this robot has sufficient experience to address service problems."
“我相信这个机器人有足够的经验来解决服务问题 。“
"I believe this robot has enough resources to solve service issues."
“我相信这个机器人有足够的资源来解决服务问题 。“
"I believe this robot can reliably resolve customer problems."
“我相信这款机器人可以可靠地解决客户的问题 。“
"I trust the information provided by this robot."
“我相信这个机器人提供的信息 。“
The emotional scale, adapted from Gursoy et al.[59], included dimensions such as sad-happy, boring-interesting, disappointed-satisfied, dissatisfied-satisfied, and angry-happy (with 1 representing negative emotions on the left and 7 denoting positive emotions on the right).
情感量表，改编自 Gursoy 等人。[59]，包括悲伤-快乐、无聊-有趣、失望-满意、不满意-满意和愤怒-快乐等维度（1 代表左侧的负面情绪，7 表示右侧的积极情绪）。
4.2.3 Data Analysis
4.2.3 数据分析
Manipulation Check
作检查
The results of the independent samples t-test showed a significant difference between the groups in terms of perceived empathy in the AI robot's responses. The high-empathetic response group rated the robot's empathy significantly higher than the low-empathetic response group (M_low empathy = 4.174, SD = 1.039 vs. M_high empathy = 5.133, SD = 0.989; t(204) = 6.786, p < 0.001).
独立样本 t 检验的 r 结果显示，在 AI 机器人反应的感知同理心方面，两组之间存在显着差异 。高移情反应组对机器人的同理心评分明显高于低移情反应组（M_low 同理心 = 4.174，SD = 1.039 vs. M_high 同理心 = 5.133，SD = 0.989;t（204） = 6.786，p < 0.001）。
The difference in perceived service failure severity was not significant (M_low empathy = 4.110, SD = 1.392 vs. M_high empathy = 3.910, SD = 1.248; t(204) = 1.112, p = 0.267), thereby ruling out any potential impact of service failure severity on the experiment. The situational stimuli were effective, and the experimental manipulation was successful.
感知到的服务故障严重程度的差异并不显著（M_low 同理心 = 4.110，SD = 1.392 vs. M_high 同理心 = 3.910，SD = 1.248;t（204） = 1.112，p = 0.267），从而排除了服务故障严重性对实验的任何潜在影响。情境刺激是有效的，实验作是成功的。
Main Effect Assessment
主效评价
The independent samples t-test was conducted with empathetic responses as the independent variable and consumer forgiveness as the dependent variable. The results indicated that, in the context of AI robot service failures, the high-empathetic response group showed significantly higher forgiveness willingness compared to the low-empathetic response group (M_low empathy = 4.448, SD = 1.078 vs. M_high empathy = 5.612, SD = 0.746; t(172.537) = 8.952, p < 0.001), confirming H1.
以移情反应 s 为自变量，消费者宽恕为因变量进行独立样本 t 检验。结果表明，在 AI 机器人服务失败的背景下 ，与低移情反应组相比，高移情反应组表现出显著更高的宽恕意愿 （M_low 同理心 = 4.448，SD = 1.078 vs. M_high 同理心 = 5.612，SD = 0.746;t（172.537） = 8.952，p < 0.001），证实 H1。
Mediation Effect Analysis
中介效应分析
The independent samples t-test revealed significant group differences in both trust and emotions. The high-empathetic response group exhibited evidently higher trust in AI robots than the low-empathetic response group (M_low empathy = 4.158, SD = 1.082 vs. M_high empathy = 5.252, SD = 0.919; t(204) = 7.843, p < 0.001), supporting H2a. Additionally, the high-empathetic group experienced substantially more positive emotions (M_low empathy = 4.066, SD = 1.252 vs. M_high empathy = 4.873, SD = 1.052; t(204) = 5.016, p < 0.001), validating H3a.
independent 样本 t 检验显示信任和情绪的组间差异显著。高移情反应组对 AI 机器人的信任度明显高于低移情反应组 （M_low 同理心 = 4.158，SD = 1.082 vs. M_high 同理心 = 5.252，SD = 0.919;t（204） = 7.843，p < 0.001），支持 H2a。此外， 高移情组经历了更多的积极情绪 （M_low 同理心 = 4.066，SD = 1.252 vs. M_high 同理心 = 4.873，SD = 1.052;t（204） = 5.016，p < 0.001）， 验证 H3a。
Using the Bootstrap method (PROCESS, Model 4, 5000 samples, 95% CI), the mediation analysis showed that both trust (LLCI = 0.304, ULCI = 0.712, excluding 0) and emotions (LLCI = 0.036, ULCI = 0.313, excluding 0) had significant mediation effects. After controlling for the mediators, the direct effect of AI robots' empathetic responses on consumer forgiveness remained significant (LLCI = 0.298, ULCI = 0.735, excluding 0), implying that trust and emotions partially mediated this relationship. Therefore, both H2b and H3b are supported.
使用 Bootstrap 方法 （PROCESS，模型 4,5000 个样本，95% CI），中介分析显示信任 （LLCI = 0.304，ULCI = 0.712，不包括 0） 和情绪 （LLCI = 0.036，ULCI = 0.313，不包括 0） 都具有显着的中介效应。在控制了中介因素后，AI 机器人的移情反应 s 对消费者宽恕度的直接影响仍然显着 （LLCI = 0.298，ULCI = 0.735，不包括 0）， 这意味着信任和情感 s 部分中介了这种关系。因此，H2b 和 H3b 均受支持。
4.2.4 Result Discussion
4.2.4 结果讨论
Experiment 2 further supports H1 in a different context, providing stronger evidence for the conclusions of Experiment 1 and confirming the mediating role of trust and emotions. Specifically, in the scenario of AI robot service failures, the high-empathetic response group demonstrates higher trust and more positive emotions compared to the low-empathetic response group. This, in turn, enhances consumers' willingness to forgive, thereby validating H2a, H2b, H3a, and H3b.
实验 2 在不同的背景下进一步支持 s H1 ，为实验 1 的结论提供了更有力的证据，并证实了信任和情感的中介作用 。具体来说，在 AI 机器人服务失败的场景中 ，与低同理心响应组相比，高同理心响应组表现出更高的信任和更积极的情绪 。这反过来又增强了消费者的宽恕意愿，从而验证了 H2a、H2b、H3a 和 H3b。
4.3 Experiment 3
4.3 实验 3
4.3.1 Experimental Design
4.3.1 实验设计
Experiment 3 employed a 2 (empathetic responses: low vs. high) × 2 (service failure severity: low vs. high) mixed experimental design to explore the moderating effect of service failure severity on consumer forgiveness.
实验 3 采用 2 （ 移情反应 s： low vs. high） × 2 （service failure severity： low vs. high） 混合实验设计来探讨服务故障严重程度对消费者宽恕性的调节作用。
4.3.2 Pre-test
4.3.2 预测试
The pre-test was adopted to assess the manipulation effectiveness of service failure severity. The experimental scenario was adapted from Yang et al.[60].
采用预测试来评估 服务故障严重程度的纵效果。实验场景改编自 Yang 等人。[60]。
In the scenario, the customer was traveling to another city and staying at a hotel with an intelligent service system. After returning from his trip, he was hungry and wished to visit the hotel's Chinese restaurant. Unable to locate it, he approached an AI robot. In the low-severity condition, the robot guided the participant to a dessert shop near the Chinese restaurant after three minutes. In cases of high severity, the robot directed him to a Western restaurant, which was far from the Chinese restaurant and unsuitable for his preferences.
在该场景中， 客户正在前往另一个城市，入住一家拥有智能服务系统的酒店。旅行回来后 ， 他饿了 ，想去酒店的中餐馆看看。由于找不到它， 他接近了一个 AI 机器人 。在严重程度较低的情况下，机器人在 3 分钟后引导参与者前往中餐馆附近的一家甜品店。在 严重的情况下 ，机器人会将他引导到一家西餐厅，该餐厅离中餐厅很远，不适合他的喜好 。
The pre-test was conducted using an online platform (Wenjuanxing), with 122 participants (50% female, 82.2% aged 19-29). Participants were randomly assigned to the high or low service failure severity condition and asked to rate the perceived severity. The measurement item was consistent with Experiment 2: "How severe do you think this service failure was?" (Likert 7-point scale, 1 = strongly disagree, 7 = strongly agree).
预测试使用在线平台 （Wenjuanxing） 进行，共有 122 名参与者 （50% 为女性，82.2% 为 19-29 岁）。参与者被随机分配到高或低服务故障严重性条件，并要求对感知的严重性进行评分。 测量项与实验 2 一致 ：“ 您认为这次服务故障有多严重？“（ 李克特 7 分制，1 = 非常不同意，7 = 非常同意）。
The independent samples t-test confirmed that participants in the high-severity group rated the failure significantly more severe than those in the low-severity group (M_low = 3.03, SD = 0.789 vs. M_high = 5.22, SD = 0.715; t(120) = 16.010, p < 0.001). The materials satisfied the requirements for experimental manipulation.
独立样本 t 检验证实，高严重性组的参与者认为失败明显 比低严重性组的参与者更严重 （M_low = 3.03，SD = 0.789 vs. M_high = 5.22，SD = 0.715;t（120） = 16.010，p < 0.001）。这些材料满足了实验作的要求。
4.3.3 Formal Experiment
4.3.3 形式实验
The formal experiment used the same scenario as the pre-test. Following the service failure, the AI robot apologized with different levels of empathy. In the low-empathetic response condition, the robot responded: "I took the wrong route. I am sorry for the inconvenience." In the high-empathetic response condition, the robot replied: "Dear customer, I am truly sorry for the inconvenience caused. I hope the kind and lovely you can give me another chance. I will make sure to take you to the right place this time. Love you!"
正式实验使用与预测试相同的场景。 在服务失败后 ，AI 机器人以不同程度的同理心道歉。在低移情反应条件下，机器人回答说：“ 我走错了路 。 对于给您带来的不便，我深表歉意。“ 在高度同理心的反应条件下，机器人回答道：” 尊敬的客户，对于给您带来的不便，我深表歉意。希望善良可爱的你能再给我一次机会。这次我一定会带你到正确的地方。爱你！“
The formal experiment was conducted with 485 participants (47.2% female, 76.1% aged 19-29) recruited via the online platform (Wenjuanxing). Participants were randomly assigned to one of four experimental conditions (empathetic response: Low/High × service failure severity: Low/High) and asked to read the experimental materials and complete the questionnaire. The survey included measures of service failure severity, empathy, trust, emotions, consumer forgiveness, and demographic information (gender and age). The measurement items remained consistent with Experiment 2. Except for demographic details, all other items were scored on a 7-point Likert scale.
正式实验是通过在线平台 （Wenjuanxing） 招募的 485 名参与者（47.2% 为女性，76.1% 为 19-29 岁）进行的。参与者被随机分配到四种实验条件之一（ 移情反应 ：低/高 × service failure severity：低/高），并被要求阅读实验材料并完成问卷。该调查包括对服务故障严重程度、同理心、信任、情绪、消费者宽恕和人口统计信息（性别和年龄）的测量。测量项目与实验 2 保持一致 。 除人口统计细节外，所有其他 项目均 按 7 分李克特量表评分 。
4.3.4 Data Analysis
4.3.4 数据分析
Manipulation Check
作检查
The independent samples t-test confirmed significant differences in the perceived level of empathy in the AI robot's responses between groups. The high-empathetic response group rated the robot's empathy significantly higher than the low-empathetic response group (M_low empathy = 4.368, SD = 1.102 vs. M_high empathy = 5.108, SD = 0.847; t(448.504) = 8.285, p < 0.001). Similarly, there were significant differences between groups in the perception of service failure severity. Participants in the high-severity group perceived the service failure as significantly more severe than those in the low-severity group. (M_low severity = 3.91, SD = 1.384 vs. M_high severity = 6.00, SD = 0.988; t(433.677) = 19.093, p < 0.001). The situational stimuli were effective, and the experimental manipulation was successful.
independent 样本 t 检验证实了 AI 机器人在组间反应中的感知同理心水平存在显着差异 。高移情反应组对机器人的同理心的评分明显高于低移情反应组（M_low 同理心 = 4.368，SD = 1.102 vs. M_high 同理心 = 5.108，SD = 0.847;t（448.504） = 8.285，p < 0.001）。同样， 在对服务故障严重程度的感知方面，各组之间存在显著差异 。 高严重性组的参与者认为服务故障明显比低严重性组的参与者更严重 。 （M_low 严重性 = 3.91，SD = 1.384 vs. M_high 严重性 = 6.00，SD = 0.988;t（433.677） = 19.093，p < 0.001）。情境刺激是有效的，实验作是成功的。
Main Effect Test
主效测试
The independent samples t-test was conducted with empathetic responses as the independent variable and consumer forgiveness as the dependent variable. The results showed that in the context of AI robot service failures, participants in the high-empathetic response group demonstrated significantly higher forgiveness intentions than those in the low-empathetic response group (M_low empathy = 4.273, SD = 0.791 vs. M_high empathy = 4.862, SD = 1.134; t(436.500) = 6.645, p < 0.001). This finding supports H1.
以移情反应 s 为自变量，消费者宽恕为因变量进行独立样本 t 检验 。结果显示，在 AI 机器人服务失败的情况下 ，高同理心反应组的参与者表现出明显高于低同理心反应组的参与者的宽恕意向 （M_low 同理心 = 4.273，SD = 0.791 vs. M_high 同理心 = 4.862，SD = 1.134;t（436.500） = 6.645，p < 0.001）。这一发现支持 H1。
Moderating Effect Verification
调节效果验证
Given both empathetic responses and service failure severity were categorical variables, a two-way ANOVA was performed. The results presented significant main effects for both empathetic responses (F(1, 481) = 53.748, p < 0.001) and service failure severity (F(1, 481) = 110.317, p < 0.001), as well as a significant interaction effect (F(1, 481) = 39.467, p < 0.001). Specifically, in the low-severity condition, the high-empathetic response group had apparently higher forgiveness intentions than the low-empathetic response group (M_low empathy = 4.443, SD = 0.501 vs. M_high empathy = 5.501, SD = 0.644; F(1, 239) = 200.588, p < 0.001). However, when the service failure severity was high, there was no remarkable difference between the two groups (M_low empathy = 4.142, SD = 0.963 vs. M_high empathy = 4.196, SD = 1.152; F(1, 242) = 0.361, p = 0.549) (Figure 2).
鉴于移情反应和服务失败严重程度都是分类变量，因此进行了双向方差分析 。结果显示 ，移情反应 （F（1， 481） = 53.748， p < 0.001） 和服务故障严重程度 （F（1， 481） = 110.317， p < 0.001） 以及显著的交互效应 （F（1， 481） = 39.467， p < 0.001）。具体来说 ， 在低严重性情况下，高移情反应组的宽恕意向明显高于低移情反应组 （M_low 同理心 = 4.443，SD = 0.501 vs. M_high 同理心 = 5.501，SD = 0.644;F（1， 239） = 200.588， p < 0.001）.然而， 当服务故障严重程度较高时， 两组之间无显著差异 （M_low 同理心 = 4.142，SD = 0.963 vs. M_high 同理心 = 4.196，SD = 1.152;F（1， 242） = 0.361， p = 0.549）（图 2）。
Figure 2. The Impact of Empathetic Response (Low vs. High)
图 2.Empathetic Response 的影响 （低与高）
× Service Failure Severity (Low vs. High) on Consumer Forgiveness
× Consumer Forgiveness 的服务故障严重性（低与高）
Note: *** represents p < 0.001; ** denotes p < 0.01; * signifies p < 0.05.
注： 代表 p < 0.001;** 表示 p < 0.01;* 表示 p < 0.05。
For the mediator variable trust, the two-way ANOVA results revealed significant main effects of empathetic responses (F(1, 481) = 27.000, p < 0.001) and service failure severity (F(1, 481) = 219.500, p < 0.001). Moreover, the two had a significant interaction effect (F(1, 481) = 13.910, p < 0.001). Specifically, in cases of low severity, the high-empathetic response group had profoundly higher trust than the low-empathetic response group (M_low empathy = 4.555, SD = 0.951 vs. M_high empathy = 5.253, SD = 0.752; F(1, 293) = 40.179, p < 0.001). However, with high service failure severity, no significant difference in trust was observed between the two groups (M_low empathy = 3.689, SD = 0.738 vs. M_high empathy = 3.803, SD = 0.982; F(1, 242) = 1.067, p = 0.303).
对于中介变量信任，双向方差分析结果揭示了移情反应 s （F（1， 481） = 27.000， p < 0.001） 和服务故障严重程度 （F（1， 481） = 219.500， p < 0.001） 的显着主效应 。此外，两者具有显著的交互作用 （F（1， 481） = 13.910， p < 0.001）。具体来说 ， 在 低严重程度的情况下，高移情反应组的信任度明显高于低移情反应组 （M_low 同理心 = 4.555，SD = 0.951 vs. M_high 同理心 = 5.253，SD = 0.752;F（1， 293） = 40.179， p < 0.001）。 然而，由于服务故障严重程度高，两组之间没有观察到信任的显著差异 （M_low 同理心 = 3.689，SD = 0.738 vs. M_high 同理心 = 3.803，SD = 0.982;F（1， 242） = 1.067， p = 0.303）。
In terms of the mediator variable emotions, the two-way ANOVA results showed significant main effects of empathetic responses (F(1, 481) = 34.955, p < 0.001) and service failure severity (F(1, 481) = 202.394, p < 0.001), along with a significant interaction effect between the two (F(1, 481) = 13.570, p < 0.001). When the service failure severity was low, the high-empathetic response group reported significantly more positive emotions than the low-empathetic response group (M_low empathy = 4.474, SD = 0.535 vs. M_high empathy = 5.227, SD = 0.638; F(1, 239) = 97.837, p < 0.001). In the high-severity condition, no significant difference in emotions was found between the two groups (M_low empathy = 3.647, SD = 1.005 vs. M_high empathy = 3.822, SD = 1.127; F(1, 242) = 1.638, p = 0.202).
就中介变量情绪而言，双向方差分析结果显示移情反应 s （F（1， 481） = 34.955， p < 0.001） 和服务故障严重程度 （F（1， 481） = 202.394， p < 0.001） 的显著主效应，以及两者之间的显著交互效应 （F（1， 481） = 13.570， p < 0.001）。 当服务故障严重程度较低时 ，高移情反应组报告的积极情绪明显多于低移情反应组 （M_low 同理心 = 4.474，SD = 0.535 vs. M_high 同理心 = 5.227，SD = 0.638;F（1， 239） = 97.837， p < 0.001）.在高严重情况下，两组之间的情绪没有显着差异 （M_low 同理心 = 3.647，SD = 1.005 vs. M_high 同理心 = 3.822，SD = 1.127;F（1， 242） = 1.638， p = 0.202）。
Bootstrap analysis was conducted to evaluate the mediating effects of trust and emotions, as well as the moderating effect of service failure severity (PROCESS, Model 8, sample size = 5000, 95% confidence interval). The results exhibited that under high service failure severity, neither trust (LLCI = -0.037, ULCI = 0.126, including 0) nor emotions (LLCI = -0.018, ULCI = 0.102, including 0) had significant mediating effects. The direct effect of empathetic responses on consumer forgiveness was also insignificant (LLCI = -0.184, ULCI = 0.193, including 0).
进行 Bootstrap 分析以评估信任和情绪的中介作用，以及服务故障严重程度的调节作用 （PROCESS，模型 8，样本量 = 5000,95% 置信区间）。结果表明 ， 在高服务故障严重程度下，信任 （LLCI = -0.037， ULCI = 0.126， 包括 0） 和情绪 （LLCI = -0.018， ULCI = 0.102， 包括 0） 都没有显著的中介作用。 移情反应 s 对消费者宽恕的直接影响也显著 （LLCI = -0.184，ULCI = 0.193，包括 0）。
In cases of low service failure severity, both trust (LLCI = 0.128, ULCI = 0.390, excluding 0) and emotions (LLCI = 0.070, ULCI = 0.258, excluding 0) had significant mediating effects. The direct effect of empathetic response on consumer forgiveness was also significant (LLCI = 0.457, ULCI = 0.857, excluding 0). Thus, hypotheses H4a, H4b, and H4c are supported.
在 服务故障严重程度较低的情况下，信任 （LLCI = 0.128， ULCI = 0.390， 排除 0） 和情绪 （LLCI = 0.070， ULCI = 0.258， 排除 0） 均具有显著的中介作用。 移情反应对消费者宽恕的直接影响也很显著（LLCI = 0.457，ULCI = 0.857，不包括 0）。 因此，假设 H4a、H4b 和 H4c 得到支持。
4.3.5 Discussion of Results
4.3.5 结果讨论
The results of Experiment 3 demonstrate that service failure severity moderates the relationship between empathetic response and consumer forgiveness. With low service failure severity, consumers in the high-empathetic response group exhibit higher trust and more positive emotions, leading to stronger forgiveness. However, in the high service failure severity condition, the effect of empathetic response on consumer forgiveness is insignificant, and the mediating role of trust and emotions is nullified.
实验 3 的结果表明，服务故障的严重程度调节了移情反应和消费者宽恕之间的关系。 在服务故障严重程度较低的情况下，高同理心反应组的消费者表现出更高的信任和更积极的情绪，从而产生更强的宽恕感。然而，在高服务故障严重性条件下， 移情反应对消费者宽恕的影响 显著 ，信任和情感的中介作用被抵消。
Therefore, H4a, H4b, and H4c are validated.
因此，H4a、H4b 和 H4c 得到验证。
5. General Discussion
5. 一般讨论
5.1 Theoretical Implications
5.1 理论意义
This study contributes to the field of AI services by delving into consumer forgiveness following AI service failures, thereby extending research into the usage phase of AI technology. Prior studies predominantly concentrated on the technology promotion and initial use of AI systems, such as robot classification, consumer acceptance, and satisfaction evaluation. However, in the age of Service 4.0, addressing consumer forgiveness after service failures is just as crucial, yet underexplored. This research bridges this gap by exploring how AI robots can prompt consumer forgiveness in the context of service failures, extending theories of service failures and recovery from human-to-human to human-machine interactions. Moreover, it examined how AI robots can independently and effectively manage service failures, expanding their application in service settings.
本研究通过深入研究 AI 服务失败后的消费者宽恕 ，从而将研究扩展到 AI 技术的使用阶段， 从而为 AI 服务领域做出贡献 。以前的研究主要集中在人工智能系统的技术推广和初始使用上，例如机器人分类、消费者接受度和满意度评估 。然而，在服务 4.0 时代，解决服务故障后消费者的宽恕 问题同样重要，但尚未得到充分探索。本研究通过探索 AI 机器人如何在服务故障的背景下促使消费者宽恕 ，将服务故障和恢复的理论从 人与人交互扩展到人机交互， 从而弥合了这一差距。 此外 ，它还研究了 AI 机器人如何独立有效地管理服务故障 ， 并扩展了它们在服务设置中的应用。
Additionally, this article enriches the theoretical discourse on emotional dynamics in AI service contexts. By industrializing emotions, AI robots can deliver tailored responses to consumers in service failure scenarios through predefined algorithms. The study explored the impact of incorporating human-like emotions—such as empathy—into AI robots, which lack inherent emotional awareness, on consumer forgiveness. The findings offer new perspectives on autonomous service recovery by AI robots, demonstrating the feasibility of enhancing service recovery outcomes by improving the empathy capabilities of AI systems.
此外，本文 还丰富了 AI 服务上下文中情感动态的理论论述。通过将情感工业化，AI 机器人可以通过预定义的算法在服务故障场景中为消费者提供量身定制的响应 。该研究探讨了将类似人类的情感（例如同理心）融入缺乏内在情感意识的 AI 机器人中对消费者宽恕的影响。这些发现为 AI 机器人的自主服务恢复提供了新的视角 ，证明了通过提高 AI 系统的同理心能力来增强服务恢复结果的可行性。
5.2 Practical Implications
5.2 实际意义
As digital technologies advance, AI service robots are broadly incorporated into industries that directly interact with consumers. As these robots become increasingly prevalent, they have more frequent encounters with consumers, which will inevitably lead to increased service failures. Therefore, businesses must proactively address these situations. Given the current limitations of AI technology, service failures in AI-mediated interactions are unavoidable. As such, businesses should prioritize the acquisition of AI robots capable of independent and timely service recovery, minimizing the adverse impact of service failures.
随着数字技术的进步，AI 服务机器人已广泛融入与消费者直接互动的行业。随着这些机器人的日益普及， 它们与消费者的接触更加频繁，这将不可避免地导致 服务故障的增加 。因此，企业必须积极应对这些情况。鉴于 AI 技术的当前局限性， AI 介导的交互中的服务故障是不可避免的。因此，企业应优先购买能够独立及时恢复服务的 AI 机器人 ，最大限度地减少服务故障的不利影响 。
Meanwhile, developers should prioritize enhancing the emotional intelligence of AI robots, allowing them to better mimic and explore human emotions. Humans are inherently emotional beings, and emotions play a pivotal role in human decision-making. AI robots can respond to humans by capturing and interpreting human emotional signals, thus influencing human behavior.
同时，开发者应该优先考虑提高 AI 机器人的情商 ，让它们更好地模仿和探索人类的情感。人类天生就是情绪化的生物， 而 e 运动在人类决策中起着举足轻重的作用 。AI 机器人 可以通过捕获和解释人类的情感信号来响应人类 ，从而影响人类的行为 。
Moreover, during recovery, companies should assess consumers' perceived severity of service failures. The results of this study imply that the empathetic responses of AI robots are effective in fostering consumer forgiveness only when the failure is perceived as low severity. This highlights the potential for AI developers to create systems capable of analyzing facial expressions, voices, and muscle activities to gauge users' perceived severity of service failures, thereby further promoting the application of AI in service industries.
此外，在恢复过程中 ， 公司应评估消费者对服务故障的感知严重性 。这项研究的结果表明 ， 只有当失败的严重程度被认为较低时，AI 机器人的移情反应才能有效地促进消费者的宽恕。这凸显了 AI 开发人员创建能够分析面部表情、语音和肌肉活动的系统的潜力，以衡量用户感知到的服务故障严重程度 ，从而进一步促进 AI 在服务行业的应用。
6. Conclusion and Limitations
6. 结论和局限性
6.1 Research Conclusion
6.1 研究结论
Based on social exchange theory, this article elaborates on the impact of empathetic responses from AI robots on consumer forgiveness following service failures. Through three experimental scenarios, the study investigated the mediating role of trust and emotions in the "empathy-forgiveness" mechanism. The main findings are as follows:
B 基于社会交换理论 ， 他的文章详细阐述了 AI 机器人的移情反应对服务失败后消费者宽恕的影响 。通过三个实验场景，研究探讨了信任和情绪在 “同理心-宽恕”机制中的中介作用。主要发现如下：
1. Empathetic responses profoundly influence consumer forgiveness in cases of service failures of AI robots. Compared to low-empathetic responses, high-empathetic responses from AI robots are more likely to foster consumer forgiveness for service failures.
1. 在 AI 机器人服务失败的情况下 ，同理心反应会深刻影响消费者的宽恕 。与低移情反应相比 ，AI 机器人的高移情反应更有可能促进消费者对服务失败的宽恕 。
2. Trust and emotions mediate the relationship between empathetic responses and consumer forgiveness. High-empathetic responses from AI robots elevate consumers' positive perceptions of robotic commitment to safeguarding customer rights, thereby fostering greater trust and more favorable emotions and promoting consumer forgiveness.
2. 信任和情感 s 调解了同理心反应和消费者宽恕之间的关系。 AI 机器人的高同理心反应提升了消费者对机器人致力于维护客户权益的积极认知，从而培养了更大的信任和更有利的情绪 ，并促进了消费者的宽恕。
3. These effects are moderated by the perceived severity of service failures. When the severity of service failure is low, high-empathetic responses provoke higher trust, more positive emotions, and greater forgiveness in comparison to low-empathetic responses. However, in cases of high severity, the effect of empathetic responses on consumer forgiveness becomes insignificant, and the mediating role of trust and emotions disappears.
3. 这些影响会受到服务故障的感知严重性的调节 。当 服务失败的严重程度较低时，与低同理心反应相比 ，高同理心反应会激发更高的信任、更积极的情绪和更大的宽恕 。然而，在高度严重的情况下 ， 移情反应 s 对消费者宽恕的影响变得微不足道，信任和情感 s 的中介作用消失。
6.2 Limitations and Future Directions
6.2 限制和未来方向
This article reviews and summarizes the current state of research on AI services, offering valuable insights into AI service recovery strategies. However, there are some shortcomings:
他的文章回顾并总结了 AI 服务研究的现状 ， 为 AI 服务恢复策略提供了有价值的见解 。但是， 也有一些缺点 ：
1. This research only involves the effectiveness of the empathy-forgiveness mechanism in the hotel industry. However, the findings may not generalize to other sectors, such as healthcare, home services, retail, or government services, necessitating further validation.
1. 本研究仅涉及 酒店业同理心-宽恕机制的有效性 。 然而， 这些发现可能无法推广到其他行业，例如医疗保健、家庭服务、零售或政府服务 ，因此需要进一步验证 。
2. Different personality traits, such as promotion focus (which leads to more positive reactions to external stimuli) and prevention focus (which causes greater sensitivity to failure), may significantly influence consumer attitudes and behaviors. Future research should investigate how these traits moderate the effectiveness of empathetic responses from AI robots in fostering forgiveness.
2. 不同的性格特征，例如促销重点（导致对外部刺激做出更积极的反应）和预防重点（ 导致对失败更敏感），可能会显着影响消费者的态度和行为。未来的研究应该调查这些特征如何调节 AI 机器人的移情反应在培养宽恕方面的有效性。
3. The current study did not distinguish between different types of service failures (e.g., process failure versus outcome failure). The impact of varying types of failures on the effectiveness of empathetic responses in service recovery remains obscure.
3. 目前的研究没有区分不同类型的服务失败 （例如，过程失败与结果失败）。 不同类型的故障对服务恢复中移情反应有效性的影响仍然不清楚 。
Future research can explore the influence of empathetic responses on consumers' forgiveness mindset from the three aspects mentioned above, thus refining the empathy-forgiveness theoretical framework.
未来的研究可以从上述三个方面探讨移情反应对消费者宽恕心态的影响， 从而提炼移情-宽恕理论框架。