随着教育信息化和现代化发展，注重个人的个性化与专门化培养、为个体的自我实现与成长助力的自适应学习已然成为教育领域的重点研究问题和未来前进方向。自适应学习系统适应于学习者的知识状态、学习需求等特征，为其推荐恰当的学习材料，而良好的推荐需以对材料特征的合理参数化为基础。目前开发和实践的自适应学习系统往往基于知识这一单特征成长情境，但在真实学习过程中，学习者的知识特征和信息加工能力特征均会改变。而且教育教学目标也不再仅包括知识的掌握，同样注重学生的能力培养和素养发展。另外，学习者信息加工能力的变化会引起其对指导型与非指导型材料的适应性变化。因此，构建知识和信息加工能力同时增长为目标的双目标自适应学习系统，推荐恰当的参数化学习材料十分重要。

基于此，本文设计并实施了三个子研究。首先，研究一完成了知识和信息加工能力双特征成长情境下的学习过程模型、测量模型等自适应学习组件设计，解决了模拟研究情境下基于认知负荷理论的学习者双特征成长与交互情况的建模问题，并采用传统的基于强化学习算法的单目标自适应学习系统验证了设计的合理性。其次，研究二构建了知识和信息加工能力同时增长的双目标自适应学习系统，改进强化学习算法的奖励函数，并比较了双目标与单目标奖励条件下学习者双特征的学习效果，实现了更适合双特征成长情境的双目标自适应学习系统。最后，研究三实现了学习材料的指导型与非指导型特征的参数化过程，提供了一种有效链接教育教学理论与算法的特征参数化方式，并在参数化材料基础上，对双目标自适应学习系统在双特征成长情境下的学习效果进行了验证与分析，对强化学习算法的材料推荐策略、学习者学习路径与过程等进行可解释性分析。

本研究考虑了实际学习过程中常见且重要的知识与信息加工能力双特征成长情境，弥补了目前自适应学习系统研究对该情境考虑不足的问题。研究结果表明：

（1）本研究构建的双目标自适应学习系统比传统的单目标自适应学习系统更适合双特征成长情境，能够获得更佳的学习者特征成长效果。

（2）参数化的学习材料能够被强化学习算法准确识别并恰当推荐。

（3）对信息加工能力的测量误差可能带来学习效果的下降，但影响并不明显；知识和信息加工能力的测量误差可能带来重复学习已掌握的知识点、在未完全掌握先序知识点条件下无效学习后序知识点等问题。

（4）强化学习推荐策略倾向于在学习初期尝试提高学习者的信息加工能力，从而获得长期知识增长收益；对不同信息加工能力水平的学习者，强化学习推荐策略能够有差异、个性化地为其提供学习支持。

With the informatization and modernization of education, adaptive learning, which focuses on cultivating talents individually and contributes to learners’ self-growth and self-actualization, has become a key research issue and future direction in the field of education. Adaptive learning systems adapt to learners' features such as knowledge state and learning needs, thus recommending appropriate learning materials. Reasonable parameterization of material features is the basis of proper recommendations. Currently, adaptive learning is often developed under a single feature growth context, such as knowledge growth. However, both the knowledge and the information processing ability of learners will change in the actual learning process. Moreover, the current goals of education include not only the acquisition of knowledge, but also the development of learners' abilities. And information processing ability growth causes changes in learners’ adaptability to instructing and non-instructing materials. Above all, it is vital to construct a dual-goal adaptive learning system that aims at simultaneous growth of knowledge and information processing ability and recommends appropriate parameterized learning materials.

Based on this, three studies were designed and implemented in the paper. In study 1, I designed the adaptive learning components such as measurement model and learning model under the dual-feature context of knowledge and information processing ability, solved the issue of modeling the dual-feature interaction based on the cognitive load theory in the simulated study situation, and verified the feasibility of my design by constructing traditional single-goal adaptive learning system based on reinforcement learning algorithms. In study 2, I constructed a dual-goal adaptive learning system with simultaneous growth of knowledge and information processing ability. The reward function in the reinforcement learning algorithm was improved, and the learning effects of learners under dual-goal and single-goal reward conditions were compared. The issue of constructing a dual-goal adaptive learning system that is more suitable for dual-feature growth situations was solved. In study 3, the parameterization process of instructing and non-instructing features of learning materials was implemented, thus providing an effective way to link pedagogical theories and algorithms. On the basis of parameterized materials, the effect of dual-goal adaptive learning system in dual-feature growth contexts was verified and analyzed. And I conducted the interpretable analysis of the material recommendation strategies based on reinforcement learning, learners' learning paths and processes, etc.

The paper considers the dual-feature growth context of knowledge and information processing ability which is common and important in the actual learning process, thus remedies the lack of consideration of this context in current adaptive learning system researches. The results show that: 1) The constructed dual-goal adaptive learning system is more suitable for the dual-feature growth context than the traditional single-goal adaptive learning system, and can achieve better learners’ features growth. 2) The parameterized learning materials can be accurately identified and appropriately recommended by the reinforcement learning algorithm. 3) The measuring error of information processing ability may bring about a decrease in learning effect, but the effect is not significant; the measuring error of knowledge and information processing ability may bring about problems such as repeated learning of knowledge points that have been mastered, and invalid learning of back-order knowledge points without mastery of a priori knowledge points. 4) The phased interpretability analysis of the reinforcement learning recommendation strategy reveals that the strategy tends to improve information processing ability at the early stage of learning process, so as to acquire long-term knowledge growth gain. For learners with different levels of information processing ability, the reinforcement learning recommendation strategy can provide learning personalized and specialized support.