Yanting Li

Yanting Li

Auto-generated profile, based on publicly available information

Yanting Li

College of Computer Science, Sichuan University, Chengdu, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
10 September 2025 in Expert Systems with Applications
Verified
Journal Article
Closed access
YLYanting Li
Yanting Li
TZTao Zhou
Tao Zhou
TCTianyu Cai
Tianyu Cai
SJShenggen Ju
Shenggen Ju

10 March 2026 in Knowledge-Based Systems
Verified
Journal Article
Closed access
TZTao Zhou
Tao Zhou
GLGuilin Li
Guilin Li
YLYanting Li
Yanting Li
RZRenyue Zhang
Renyue Zhang
SJShenggen Ju
Shenggen Ju

10 November 2025
Verified
Conference Paper
Open access
TCTianyu Cai
Tianyu Cai
XHXiaodi Huang
Xiaodi Huang
TZTao Zhou
Tao Zhou
YLYanting Li
Yanting Li
SJShenggen Ju
Shenggen Ju

21 April 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
YLYanting Li
Yanting Li
SJShenggen Ju
Shenggen Ju
TZTao Zhou
Tao Zhou
XYXia Yang
Xia Yang
Knowledge tracing (KT) models students' knowledge states to predict their future performance in intelligent tutoring systems. Despite recent progress, existing methods suffer from two key limitations: first, they predominantly model knowledge states at a single timescale, failing to simultaneously capture long-term cognitive trends and short-term response fluctuations. Second, they typically fuse multidimensional learning behavior features into a unified representation space, thereby failing to disentangle the individual contributions of different behavioral factors. To address these limitations, we propose a Dual-Decoupled Multi-scale Knowledge Tracing model (DDKT) that decouples the learning process across behavioral and temporal dimensions. In the behavioral dimension, a neural additive behavior gating mechanism explicitly characterizes the modulatory effect of multidimensional behavior features on the intensity of knowledge state updates. In the temporal dimension, a bottom-up guided trend pyramid structure and multi-scale parallel depthwise separable convolutional branches model long-term cognitive trends and short-term response fluctuations, respectively. Furthermore, a skill-aware forgetting module and a FiLM-based prediction module are introduced to capture heterogeneous forgetting patterns across knowledge concepts while enhancing model interpretability. Experimental results on four widely used public datasets demonstrate that DDKT consistently outperforms 20 representative state-of-the-art KT methods across multiple evaluation metrics. Further experiments and analyses confirm that DDKT enhances model interpretability while maintaining competitive predictive performance, indicating its potential applicability in intelligent educational scenarios. The code of our method is available at https://anonymous.4open.science/r/DDKT-DFC5.