This paper proposes extended Long Short-Term Memory (LSTM) networks for the knowledge tracing task and employs explainable AI methods to address interpretability issues. Specifically, we developed an extended LSTM-based model to automatically diagnose students' knowledge states. We then leveraged three interpreting methods—gradient sensitivity, gradient*input, and Deep SHAP—to explain the model's predictions by computing input contributions. The results demonstrate that the proposed model outperforms DKT, and the three methods effectively explain its predictions. Additionally, we identified three key insights into the model's working mechanisms.

Artificial intelligence (AI) has been employed to provide automated analysis of collaborative argumentation due to its importance. However, traditional deep learning models face challenges with generalizability to other dimensions and contexts. Existing studies on large language models (LLMs) for classroom dialogue primarily rely on prompt engineering techniques because of the high costs associated with fully fine-tuning LLMs. This approach results in limited performance, indicating a need for improvement. To address these issues, this study proposes the use of parameter-efficient fine-tuning (PEFT) techniques to optimize the performance of LLMs in analyzing classroom collaborative argumentation. Specifically, we utilized Low-Rank Adaptation (LoRA), a well-known PEFT technique, to fine-tune two state-of-the-art LLMs, Llama-3.2-3B and Gemma-2-9B. The results demonstrate that, compared to fully fine-tuning BERT and RoBERTa, using LoRA for PEFT of Llama-3.2-3B and Gemma-2-9B achieves superior performance in analyzing argument moves within collaborative argumentation. We conclude that PEFT techniques provide a promising direction for classroom dialogue analysis.

In e-learning, extracting suitable learning objects (LOs) from a vast resource pool and organizing them into high-quality learning paths is crucial for helping e-learners achieve their goals. Numerous approaches have been proposed to recommend optimal learning paths for e-learners. However, it is essential to emphasize that e-learning systems typically consist of a wide range of LOs with varying levels of granularity, ranging from fine-grained to coarse-grained. Unfortunately, current research has not adequately considered the underlying granularity structure of LOs when optimizing learning paths. Existing methods primarily focus on organizing LOs at a single granularity level, limiting their applicability in real-world e-learning systems. To address the limitations, we propose a multigranularity learning path recommendation (MGLPR) framework that aims to flexibly and effectively integrate the diverse granularity levels of LOs into high-quality learning paths. In this framework, a two-layer [knowledge point (KP) and LO layers] model is developed to formulate the MGLPR problem as a constrained optimization problem and an improved ant colony optimization algorithm (IACO) is introduced to solve it to identify optimal learning paths for e-learners. To evaluate the effectiveness of the proposed IACO, we conducted extensive computational experiments using 30 simulation datasets with varying problem sizes and complexities. The results demonstrate that our proposed IACO achieves superior performance and robustness compared with other competitors. Additionally, an empirical study was conducted to investigate the efficacy of the proposed approach in an authentic learning context, with results indicating that the proposed method outperforms the traditional self-organized ones.

This paper explores proposing interpreting methods from explainable artificial intelligence to address the interpretability issues in deep learning-based models for classroom dialogue. Specifically, we developed a Bert-based model to automatically detect student talk moves within classroom dialogues, utilizing the TalkMoves dataset. Subsequently, we proposed three generic interpreting methods, namely saliency, input*gradient, and integrated gradient, to explain the predictions of classroom dialogue models by computing input relevance (i.e., contribution). The experimental results show that the three interpreting methods can effectively unravel the classroom dialogue analysis, thereby potentially fostering teachers' trust.

Deep learning-based knowledge tracing (DLKT) models have been regarded as the promising solution to estimate learners’ knowledge states and predict their future performance based on historical exercise records. However, the increasing complexity and diversity make DLKT models still difficult for users, typically including both learners and teachers, to understand models’ estimation results, directly hindering the model’s deployment and application. Previous studies have explored using methods from explainable artificial intelligence (xAI) to interpret DLKT models, but the methods have been limited in their generalizing capability and inefficient interpreting procedures. To address these limitations, we proposed a simple but efficient model-agnostic interpreting method, called Gradient*Input, to explain the predictions made by these models in two datasets. Comprehensive experiments have been conducted on the existing five DLKT models with representative neural network architectures. The experiment results showed that the method was effective in explaining the predictions of DLKT models. Further analysis of the interpreting results revealed that all five DLKT models share a similar rule in predicting learners’ item responses, and the role of skill and temporal information was found and discussed. We also suggested potential avenues for investigating the interpretability of DLKT models.

Using visual discourse tools can be a valuable approach for teachers to foster academically productive talk in the classroom. However, teachers often pay considerable attention to highly engaged students and fail to create sufficient opportunities for less engaged students to participate in classroom dialogue. This study seeks to enhance learning by incorporating social network analysis to elucidate students’ levels of engagement in classroom discourse. Over a three-week period, this study analyzed both overall and individual student’s dialogic engagement in the classroom. The results demonstrate that as teachers purposefully lead class dialogue and limit their own speech, students gradually speak more. Additionally, those who initially speak less are given more opportunities to engage in classroom dialogue. These results emphasize the significance of using visualization tools to assist teachers in orchestrating and optimizing classroom dialogue.

The recommendation of personalized learning paths is recognized as one of the most challenging aspects in the field of e-learning. In the existing literature, numerous approaches have been proposed to identify appropriate learning paths for e-learners, taking into consideration multiple perspectives. However, the current state of research lacks a unified framework that effectively integrates the most vital parameters associated with the learner, learning object (LO), and domain knowledge to generate optimal learning paths. To address this challenge, a novel bio-inspired approach is proposed for solving the personalized learning path problem. In this method, we initially incorporate the learner, LO and domain knowledge models into a unified mathematical model. Then an enhanced ant colony optimization algorithm is utilized to determine the optimal personalized learning paths for learners. To investigate the effectiveness of the proposed method, we performed several computational experiments based on six simulation datasets. The results indicate that the proposed method surpasses other competing methods in terms of performance and robustness, showcasing its superior effectiveness.