DeepSeek使用强度与大学生学习适应——基于遮掩效应与归因方式的实证研究

DeepSeek Usage Intensity and University Students’ Learning Adaptation—An Empirical Study Based on the Masking Effect and Attributional Styles Scenarios

摘要: 在生成式人工智能深度渗透高等教育的背景下，本研究聚焦DeepSeek使用强度与大学生学习适应之间的复杂关系，引入AI自我效能感作为遮掩变量、归因方式作为个体差异变量，构建“使用强度→自我效能感→学习适应”的遮掩效应模型。以514名使用DeepSeek的大学生为样本，采用问卷调查与分层回归分析，结果表明：(1) DeepSeek使用强度对学习适应具有显著负向直接效应，验证了技术替代效应的存在；(2) AI自我效能感在二者之间起遮掩中介作用，间接效应(正向)与直接效应(负向)符号相反，控制中介变量后负向效应增强，揭示自我效能感既缓冲技术替代冲击、又掩盖过度使用危害的双重属性；(3) 归因方式的组间比较显示，内部归因者表现出显著更高的使用强度与自我效能感，但学习适应水平与外部归因者无显著差异，呈现“高使用–高自信–未发现显著优势”现象。研究拓展了技术接受模型与自我效能感理论在AI教育应用情境中的解释边界，为高校差异化AI素养教育提供了实证依据与干预靶点。

Abstract: Against the backdrop of generative artificial intelligence’s deep penetration into higher education, this study investigates the complex relationship between DeepSeek usage intensity and college students’ learning adaptation, introducing AI self-efficacy as a suppressing variable and attribution style as an individual-difference variable. A suppression effect model of “usage intensity → self-efficacy → learning adaptation” was constructed and tested with a sample of 514 college students who use DeepSeek, employing questionnaire surveys and hierarchical regression analysis. The results indicate that: (1) DeepSeek usage intensity has a significant negative direct effect on learning adaptation, confirming the existence of a technology substitution effect; (2) AI self-efficacy serves as a significant suppressor in the relationship between usage intensity and learning adaptation, with the indirect effect (positive) opposite in sign to the direct effect (negative), and the negative direct effect strengthening after controlling for the mediator, revealing self-efficacy’s dual nature of both buffering the negative impact of technology substitution and masking the hidden costs of overuse; (3) between-group comparisons by attribution style show that internal attributors exhibit significantly higher usage intensity and self-efficacy, yet demonstrate no advantage in learning adaptation compared to external attributors, presenting a “high usage-high confidence-no significance advantages found” phenomenon. This study extends the explanatory boundaries of the Technology Acceptance Model and self-efficacy theory within AI educational contexts, providing empirical evidence and intervention targets for differentiated AI literacy education in higher education institutions.