Intelligent education systems have enabled personalized learning (PL). In PL, students are presented with educational contents that are consistent with their personal knowledge states (KS), and the critical task is accurately estimating these states through data. Knowledge tracing (KT) infers KS (latent) through historical student interactions (observed) with the knowledge components (KCs). A wide variety of KT techniques have been developed, from Bayesian Knowledge Tracing (BKT) to Deep Knowledge Tracing (DKT). However, in most of these methods, the KCs are represented as stand-alone entities, and the effect of representing KCs using contexts such as learning-related factors has been under-investigated. Also, KT needs to generate personalized results to facilitate tasks such as exercise recommendation. In this paper, we propose two approaches that use a contextualized representation of KCs, one with a content-based approach and another with a Long Short Term Memory (LSTM) network plus a personalization mechanism. By performing extensive experiments on two real-world datasets, results show not only a tangible improvement in prediction accuracy in the KT task compared to existing methods, but also its effectiveness in improving the recommendation precision.

智能教育系统已启用个性化学习（PL）。在PL中，向学生展示与他们的个人知识状态（KS）相一致的教育内容，而关键任务是通过数据准确估计这些状态。知识跟踪（KT）通过历史学生与知识组件（KC）的交互（观察到）来推断KS（潜在）。从贝叶斯知识跟踪（BKT）到深度知识跟踪（DKT），已经开发了各种各样的KT技术。但是，在大多数这些方法中，KC均表示为独立实体，并且使用诸如学习相关因素之类的上下文表示KC的效果尚未得到充分研究。此外，KT需要生成个性化的结果以促进诸如锻炼推荐之类的任务。在本文中，我们提出了两种使用KC的上下文表示的方法，一种是基于内容的方法，另一种是使用长短期记忆（LSTM）网络以及个性化机制。通过在两个真实世界的数据集上进行广泛的实验，结果不仅表明与现有方法相比，KT任务的预测准确性有了明显的提高，而且还提高了推荐精度。