人工智能赋能乡村教育的县域研究：困境与突破路径

County-Level Research on AI-Empowered Rural Education: Dilemmas and Breakthrough Path

摘要: 城乡教育差距是我国基础教育均衡发展面临的核心挑战之一。县域乡村学校普遍存在师资力量薄弱、优质教学资源匮乏、难以开展个性化教学和家校共育薄弱等问题。本文以川西南某县域3所乡镇中心小学、2所村级教学点的实践项目为基础，采用分层整群抽样、准实验研究、问卷调查、课堂观察、深度访谈与平台日志分析相结合的方法，探讨人工智能技术在县域乡村教育中的适配路径。项目基于Dify 0.15.x、Qwen2.5-7B-Instruct、bge-m3向量模型等开源工具，设计并实现了集智能备课、个性化学习路径规划、作业自动批改与学情分析于一体的AI辅助教学平台。研究在建构主义、掌握学习和最近发展区理论基础上，构建“数据诊断–智能支持–教师裁量–反馈优化”的理论与技术框架。三个月准实验结果显示，实验组教师备课时间由182.4 ± 44.8分钟/课时降至104.7 ± 36.2分钟/课时，作业批改时间由153.6 ± 38.5分钟/班降至48.7 ± 22.4分钟/班；与对照组相比，差异中的差异检验均达到显著水平(p < 0.001)。学生层面，实验组学困生知识点巩固率提升31.7个百分点，语文、数学及格率较对照组分别提高15.2和18.7个百分点。质性资料进一步表明，平台能够减轻教师重复性劳动、改善即时反馈和资源可及性，但也存在主观题批改一致性不足、教师技术焦虑、算法推荐可能窄化学习经验等风险。研究认为，低成本开源AI平台可为“人工智能 + 乡村教育”提供可行路径，但其价值实现依赖教师主导、数据治理和可持续的区域协同机制。

Abstract: The urban-rural education gap is one of the core challenges facing the balanced development of basic education in China. County-level rural schools generally face issues such as weak teacher resources, lack of high-quality teaching resources, difficulty in implementing personalized teaching, and weak home-school collaboration. Based on a practical project involving three township central primary schools and two village-level teaching points in a certain county in southwestern Sichuan, this paper adopts a combination of stratified cluster sampling, quasi-experimental research, questionnaire surveys, classroom observations, in-depth interviews, and platform log analysis to explore the adaptive path of artificial intelligence (AI) technology in county-level rural education. The project is based on open-source tools such as Dify 0.15.x, Qwen2.5-7B-Instruct, and bge-m3 vector model, and designs and implements an AI-assisted teaching platform that integrates intelligent lesson preparation, personalized learning path planning, automatic homework grading, and learning situation analysis. Based on constructivism, mastery learning, and the zone of proximal development theory, the study constructs a theoretical and technical framework of “data diagnosis - intelligent support - teacher discretion - feedback optimization”. The three-month quasi-experimental results show that the preparation time for teachers in the experimental group decreased from 182.4 ± 44.8 minutes per class period to 104.7 ± 36.2 minutes per class period, and the homework grading time decreased from 153.6 ± 38.5 minutes per class to 48.7 ± 22.4 minutes per class; compared with the control group, the differences in the difference test reached a significant level (p < 0.001). At the student level, the knowledge consolidation rate of students with learning difficulties in the experimental group increased by 31.7 percentage points, and the passing rates for Chinese and mathematics increased by 15.2 and 18.7 percentage points, respectively, compared with the control group. Qualitative data further indicates that the platform can reduce teachers’ repetitive work, improve immediate feedback and resource accessibility, but there are also risks such as insufficient consistency in grading subjective questions, teacher technology anxiety, and the potential narrowing of learning experiences due to algorithm recommendations. The study believes that low-cost open-source AI platforms can provide a feasible path for “AI + rural education”, but the realization of its value depends on teacher-led initiatives, data governance, and sustainable regional collaboration mechanisms.