Dual-State Personalized Knowledge Tracing with Emotional Incorporation

TL;DR

This work introduces DEKT, a dual-state personalized KT framework that explicitly incorporates student emotions into learning dynamics. It features a Knowledge State Boosting Module that uses Emotion-Boosted learning gains and an Emotional State Tracing Module to predict emotional representations from personalized emotional states, feeding into a Comprehensive Predicting Module for emotion-informed response prediction. A transfer-enabled variant, T-DEKT, extends applicability to emotion-sparse datasets by autoregressively predicting emotions and reusing emotion embeddings from related data. Extensive experiments on ASSISTments and EdNet-KT1 demonstrate state-of-the-art performance and strong generalization, with ablations confirming the critical role of emotion-driven gains and emotion prediction. This approach offers a principled pathway to more accurate, personalized, and transferable KT in real-world online learning systems.

Abstract

Knowledge tracing has been widely used in online learning systems to guide the students' future learning. However, most existing KT models primarily focus on extracting abundant information from the question sets and explore the relationships between them, but ignore the personalized student behavioral information in the learning process. This will limit the model's ability to accurately capture the personalized knowledge states of students and reasonably predict their performances. To alleviate this limitation, we explicitly models the personalized learning process by incorporating the emotions, a representative personalized behavior in the learning process, into KT framework. Specifically, we present a novel Dual-State Personalized Knowledge Tracing with Emotional Incorporation model to achieve this goal: Firstly, we incorporate emotional information into the modeling process of knowledge state, resulting in the Knowledge State Boosting Module. Secondly, we design an Emotional State Tracing Module to monitor students' personalized emotional states, and propose an emotion prediction method based on personalized emotional states. Finally, we apply the predicted emotions to enhance students' response prediction. Furthermore, to extend the generalization capability of our model across different datasets, we design a transferred version of DEKT, named Transfer Learning-based Self-loop model (T-DEKT). Extensive experiments show our method achieves the state-of-the-art performance.

Figures (7)

• Figure 1: (a) presents the data information of two emotional attributes in the ASSISTchall dataset. The bar chart illustrates the distribution of emotions for each interval with a size of 0.1, while the line chart depicts the corresponding answer accuracy within each interval. (b) presents the changes in four types of emotions during the completion of the same sequence of questions for two students with different inherent traits, Tom and Joe. The four emotions referred to as concentration, boredom, confusion and frustration, respectively.
• Figure 2: The changes in the personalized emotional state during a student's answering process.
• Figure 3: The main structure of our DEKT model. We show the processing pipeline of DEKT at step $t$. In this timestep, we model two parallel processes. In the first process, the inputs are the exercise $e_t$, the answer time $at_t$, and the student's answer $a_t$, and the output is the knowledge state $h_t$ after answering. In the second process, the inputs are the exercise $e_t$, and the emotional embedding ${cm}_t$, and the output is the emotional state $f_t$ after answering. Moreover, we will predict the student’s performance $y_{t+1}$ at timestep $t+1$.
• Figure 4: The architecture of T-DEKT model
• Figure 5: Illustrates the dynamic changes in the knowledge state $h_t$ and emotional state $f_t$ of a student throughout the process of consecutively solving 13 exercises involving 3 different concepts.