Graded reading is one of the important ways of English learning. How to automatically judge and grade the difficulty of English reading corpus is of great significance for precision teaching and personalized learning. However, the current rule-based readability assessment methods have some limitations, such as low efficiency and poor accuracy. Specially, these traditional methods usually lack semantics that is crucial for students to understand the reading material. Meanwhile, they are difficult to be mapped to the difficulty level, which is not conducive to flexible application in actual personalized teaching. In this study, a method for grading the difficulty of English reading corpus is proposed. This approach makes use of a pre-trained language model and feature fusion embedding to make the most of multi-feature data when training. First, based on linguists' evaluations of the variables influencing the difficulty of English reading corpus, three primary statistical features—sentence length, word length, and the number of prepositions—are taken into consideration. Then, the semantic features and part-of-speech features of the text are learned by pre-trained language model and LSTM respectively to capture polysemy features and fine-grained semantic representations that are difficult to represent with traditional models. Finally, multi-feature embedding extractions are fused to grade the difficulty of English reading corpus. Extensive experimental comparisons on a self-built dataset and two datasets that are freely accessible with various models indicate that our method outperform the others in the task of grading the difficulty of English reading corpora.

With the rapid development of deep recommendation models, Multitask Learning (MTL) enables models to have better generalization ability in the learning process by extracting relevant information from different tasks. Although many studies have focused on extracting task-specific information from expert networks using attention mechanisms, they have failed to fully exploit the rich information among task towers, resulting in the failure of each task to use the information to improve the model's effectiveness. In this paper, design a Cross and Mixture (CM) unit for multitask recommendation to improve the feature exploitation capability of models through feature interaction between task towers. The results show that adding this feature fusion method to the existing multitask model higher the original effect of three common MTL datasets.

Knowledge tracing aims to predict students’ future performance based on their historical learning interactions, thereby supporting learner modeling in online learning platforms and intelligent tutoring systems. Recent deep learning-based knowledge tracing models have achieved strong predictive performance, but many of them rely on increasingly complex architectures or additional optimization objectives, which may increase training difficulty and computational cost. To examine whether recurrent models can remain competitive when equipped with more effective exercise representation, this paper proposes ExerCAKT, an Exercise Context-Aware Knowledge Tracing model based on gated recurrent units. Different from conventional GRU-based knowledge tracing models that mainly use interaction sequences to update students’ knowledge states, ExerCAKT explicitly separates knowledge-state modeling and exercise-context modeling through two GRU-based feature extractors. The knowledge-state feature extractor captures students’ evolving mastery patterns, while the exercise feature extractor models contextual exercise information to enhance question-level prediction without introducing extra optimization objectives. Experiments on four public knowledge tracing datasets show that ExerCAKT achieves the best AUC results in six out of seven evaluation settings and obtains similar advantages in ACC. At the question level, ExerCAKT improves AUC by 2.00–4.71% over DKT on AL2005, AS2009, and NIPS34. At the knowledge-concept level on AS2009, AL2005, and NIPS34, it improves AUC by 0.24–1.95% over AKT. These results suggest that recurrent knowledge tracing models can still achieve competitive performance when exercise contextualization is explicitly modeled.