DeepSeek在高考志愿填报中的应用研究
A Study on Applying DeepSeek to Gaokao University Application
摘要
高考志愿填报是关乎考生切身利益的重要决策环节，长期存在信息不对称、决策维度复杂、咨询需求高度个性化等问题。随着大语言模型在自然语言理解与生成方面取得突破性进展，将其引入高考志愿填报咨询场景具有显著的应用价值。本文以DeepSeek系列大模型为技术底座，设计并实现了面向高考志愿填报的智能问答系统。首先，整合高校招生政策、专业设置、历年录取分数线与就业前景等多源异构数据，构建高考志愿填报领域知识图谱；其次，基于DeepSeek的多头潜在注意力（MLA）、多词元预测（MTP）与混合专家（MoE）等核心技术，实现用户意图精准识别、多轮对话管理与个性化志愿推荐；最后，引入录取概率推理与滑档、退档风险预警机制，提升填报方案的科学性。在原型系统上的试点测试表明，所提方法在意图识别准确率、回答准确率与用户满意度方面均优于传统规则式与检索式问答系统，能够为考生与家长提供高效、可靠的决策支持。研究结果可为教育决策支持系统的智能化升级提供理论参考与技术借鉴。
Abstract
Filling out university applications for the National College Entrance Examination (Gaokao) is a critical decision-making process that directly affects examinees\' interests. It has long been plagued by issues such as information asymmetry, multi-dimensional decision complexity, and highly individualized consultation demands. With breakthrough advances of large language models in natural language understanding and generation, introducing such models into Gaokao university application consultation scenarios delivers notable application value. This paper designs and implements an intelligent question-and-answer system for Gaokao university application, using the DeepSeek series large models as the technical foundation.First, multi-source heterogeneous data including university admission policies, program offerings, historical admission cut-off scores and employment prospects are integrated to construct a domain knowledge graph for Gaokao university application. Second, leveraging core technologies of DeepSeek such as Multi-head Latent Attention (MLA), Multi-Token Prediction (MTP), and Mixture of Experts (MoE), the system realizes accurate user intent recognition, multi-turn dialogue management and personalized university major recommendation. Finally, admission probability inference together with early warning mechanisms for filing slippage and file rejection risks are incorporated to improve the rationality of application plans.Pilot tests on the prototype system demonstrate that the proposed approach outperforms traditional rule-based and retrieval-based Q&A systems in intent recognition accuracy, answer accuracy and user satisfaction. It can provide efficient and reliable decision support for examinees and their parents. The research findings offer theoretical references and technical insights for the intelligent upgrade of educational decision support systems.关键词
DeepSeek/高考志愿填报/智能问答/知识图谱/大语言模型Key words
DeepSeek/college entrance examination/intelligent question answering/knowledge graph/large language model引用本文复制引用
钟增胜.DeepSeek在高考志愿填报中的应用研究[EB/OL].(2026-09-29)[2026-10-01].http://www.paper.edu.cn/releasepaper/content/202609-34.学科分类
教育
