联系我们

论文查询结果	返回搜索

论文编号：	16132
作者编号：	2120243790
上传时间：	2026/6/8 22:37:24
中文题目：	AI 购买建议对电商导购平台中消费者参与影响研究
英文题目：	Research on the Impact of AI Purchase Recommendations on Consumer Participation in E-commerce Shopping Guide Platforms
指导老师：	李凯
中文关键字：	AI 购买建议；内容型电商社区；消费者参与行为；信息结构；交易情 境；情感表达强度；信息诊断性；电子口碑
英文关键字：	Key Words: AI purchase advice; Content-based e-commerce community; Consumer engagement behavior; Information structure; Transaction context; Emotional expression intensity; Information diagnosticity; Electronic word of mouth
中文摘要：	摘要 近年来，生成式人工智能正以“购买建议”“要点总结”等前台内容形态进入内容型电商社区，电商平台通过自动生成结构化的文本参与消费者的信息决策过程。在这一背景下，生成式人工智能不仅改变了信息生产效率，也在通过影响用户在社区中的参与行为，进而影响内容传播、平台互动生态与营销绩效。现有研究多从信任、采纳意向等心理变量切入，或聚焦推荐排序等后台算法机制，对于生成式人工智能建议在真实平台中能否提升消费者参与，以及其效果如何受到信息结构与情境因素影响，仍缺乏基于自然场景行为数据的系统证据。营销传播研究长期强调信息呈现结构的重要性，但在生成式人工智能语境下，“AI怎么说”是否会改变消费者参与的类型与强度，仍有待进一步检验。 基于此，本文以“什么值得买”平台为研究场景，围绕生成式人工智能购买建议对消费者参与行为的影响展开实证研究。本文以评论量与收藏量作为核心因变量，并对两项指标进行对数化处理，以刻画不同类型的消费者参与行为：其中，评论是一种高投入的互动行为，代表着贡献型参与；收藏是一种低投入互动行为，代表留存型参与。研究首先比较帖子是否展示AI购买建议所带来的总体差异；进一步在只含有AI建议的样本中，将AI文本按照信息结构划分为单边信息与双边信息两种类型，并检验二者对参与行为的差异化影响；在此基础上，引入交易情境变量，检验价格水平与折扣强度对AI效应的边界作用；同时，结合AI文本的情感表达强度，检验表达风格对信息结构效应的调节作用；最后，本文通过替换计数模型设定与倾向得分匹配方法提升估计结果的稳健性，并引入信息诊断性机制变量开展中介检验，以揭示生成式人工智能购买建议影响消费者参与行为的潜在路径。 实证结果显示，相较于未展示AI购买建议的帖子，展示AI购买建议的帖子在评论与收藏两个维度上均呈现显著更高的参与水平，说明生成式AI购买建议整体具有参与促进效应。进一步的情境边界检验表明，AI购买建议的促进效应并非在所有交易情境下相同：随着商品价格水平上升以及折扣力度加深，AI购买建议对消费者参与的边际促进作用显著减弱，表明高涉入与高不确定的交易场景会压缩生成式AI建议的增量价值。与此同时，在仅包含AI购买建议的样本中，信息结构效应呈现更精细的差异：双边信息相较单边信息在收藏维度上表现出更稳定的正向优势，而在评论维度上的平均差异并不稳定，但在价格更高的场景中，双边信息在评论维度的相对优势显著增强，说明双边结构更可能在高风险决策语境下被用户用于进一步互动与信息验证。 机制检验进一步揭示，信息诊断性在双边信息影响参与行为的过程中发挥部分中介作用。双边信息显著提升AI建议文本所提供的信息诊断性水平，而信息诊断性又显著促进消费者参与；在引入机制变量后，双边信息对参与行为的直接效应有所下降但仍保持显著，表明双边结构通过提升信息可用于判断的程度增强参与。表达风格方面，情感表达强度对信息结构效应形成关键约束：在情感表达较弱的情况下，双边信息对评论与收藏具有显著促进作用；当情感表达处于中等强度时，双边信息的相对优势被显著削弱，表明更强的情感色彩可能引发用户对文本立场与说服意图的敏感，从而降低双边结构所带来的客观性与可信优势。 稳健性检验同样支持上述结论。基于Poisson与负二项计数模型的估计结果与基准回归一致，且在AI样本内部采用倾向得分匹配方法缓解结构类型选择偏误后，双边信息相较单边信息对评论与收藏的平均处理效应仍显著为正，说明研究结论具有较强稳健性。 总体而言，本文表明生成式AI购买建议在内容型电商社区中能够提升消费者参与，但其效果受到交易情境约束，并且“信息结构”与“情感表达”共同决定生成式AI建议能否将信息增量转化为可观测的用户参与。本文的主要贡献在于，将营销传播中的单边与双边信息理论拓展至可见生成式AI购买建议场景，基于真实平台行为数据同时刻画总体效应、边界条件与机制路径，为平台AIGC模块的内容策略优化与治理提供经验证据。 关键词：AI购买建议；内容型电商社区；消费者参与行为；信息结构；交易情境；情感表达强度；信息诊断性；电子口碑
英文摘要：	Abstract In recent years, generative artificial intelligence (AI) has entered content-based e-commerce communities in visible formats such as “purchase advice” and “key-point summaries.” By automatically generating structured text, platforms directly participate in consumers’ information processing and decision-making. In this context, generative AI not only improves the efficiency of information production, but may also influence user engagement in the community, thereby shaping content diffusion, the interaction ecosystem, and marketing performance. Prior studies mostly focus on psychological constructs such as trust and adoption intention, or examine back-end algorithmic mechanisms such as ranking and recommendation. However, systematic evidence based on real-world behavioral data is still limited regarding whether visible generative AI advice can increase consumer engagement on actual platforms, and how its effects depend on information structure and contextual factors. Marketing communication research has long emphasized the role of message structure, yet under a generative AI setting, whether “how AI speaks” changes the type and intensity of consumer engagement remains to be tested. Against this background, this study uses the “SMZDM ” platform as the research setting and conducts an empirical analysis of how generative AI purchase advice affects consumer engagement behaviors. Comment volume and favorite volume are used as the key dependent variables, and both measures are log-transformed to capture different types of engagement. Specifically, commenting is a high-effort interactive behavior that reflects contributive engagement, whereas favoriting is a low-effort interactive behavior that reflects retention-oriented engagement. The analysis first compares the overall difference between posts with and without displayed AI purchase advice. Next, within the subsample that contains AI advice, AI-generated text is classified into one-sided versus two-sided information structures, and their differential effects on engagement are examined. Building on this, transaction-context variables are introduced to test boundary conditions, focusing on product price level and discount depth. In addition, the emotional expression intensity of AI text is measured to examine whether expression style moderates the information-structure effect. Finally, robustness is assessed by alternative count-model specifications and propensity score matching, and a mediation test is conducted by introducing information diagnosticity as the mechanism variable to uncover the potential pathway through which generative AI purchase advice influences consumer engagement. The empirical results show that, compared with posts without displayed AI purchase advice, posts with AI purchase advice exhibit significantly higher levels of engagement in both comments and favorites, indicating an overall engagement-promoting effect of generative AI purchase advice. Further boundary analyses suggest that this effect is not uniform across transaction contexts: as product price increases and discounts become deeper, the marginal engagement effect of AI purchase advice significantly weakens. This finding implies that high-involvement and high-uncertainty purchasing contexts compress the incremental value of generative AI advice. Meanwhile, within the AI-advice subsample, the information-structure effect reveals more nuanced patterns. Two-sided information, relative to one-sided information, shows a more stable positive advantage in favoriting, while the average difference in commenting is not stable. However, in higher-price contexts, the relative advantage of two-sided information in commenting becomes significantly stronger, suggesting that two-sided structures are more likely to be used for further interaction and information verification in high-risk decision settings. Mechanism tests further indicate that information diagnosticity plays a partial mediating role in the effect of two-sided information on engagement. Two-sided information significantly increases the perceived diagnosticity of AI-generated advice, and information diagnosticity, in turn, significantly promotes consumer engagement. After including the mechanism variable, the direct effect of two-sided information on engagement declines but remains significant, indicating that two-sided structure enhances engagement partly by increasing the extent to which information can be used for judgment. Regarding expression style, emotional expression intensity imposes a key constraint on the information-structure effect. When emotional expression is weak, two-sided information significantly increases both comments and favorites. When emotional expression is at a moderate level, the relative advantage of two-sided information is significantly reduced, suggesting that stronger emotional tone may increase users’ sensitivity to stance and persuasive intent, thereby weakening the objectivity and credibility advantage typically associated with a two-sided structure. Robustness checks support these conclusions. Estimates based on Poisson and negative binomial count models are consistent with the baseline results. Moreover, after applying propensity score matching within the AI subsample to mitigate selection bias in structure type, the average treatment effect of two-sided information (relative to one-sided information) on comments and favorites remains significantly positive, indicating strong robustness. Overall, this study shows that generative AI purchase advice can increase consumer engagement in content-based e-commerce communities, but its effects are constrained by transaction context, and both information structure and emotional expression jointly determine whether the informational increment from generative AI can translate into observable user engagement. The main contribution of this study is to extend one-sided versus two-sided message theory in marketing communication to the setting of visible generative AI purchase advice, and to provide evidence on the overall effect, boundary conditions, and mechanism pathway using real-world platform behavioral data. The findings also offer empirical implications for optimizing content strategy and governance of platform AIGC modules. Key Words: AI purchase advice; Content-based e-commerce community; Consumer engagement behavior; Information structure; Transaction context; Emotional expression intensity; Information diagnosticity; Electronic word of mouth
查看全文：	预览  下载(下载需要进行登录)