Symbolic Cognitive Diagnosis via Hybrid Optimization for Intelligent Education Systems

Authors: Junhao Shen, Hong Qian, Wei Zhang, Aimin Zhou

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The extensive experimental results on various real-world datasets show the superiority of SCD on both generalization and interpretability. The ablation study verifies the efficacy of each ingredient in SCD, and the case study explicitly showcases how the interpretable ability of SCD works.
Researcher Affiliation Academia Junhao Shen, Hong Qian*, Wei Zhang, Aimin Zhou Shanghai Institute of AI for Education and School of Computer Science and Technology, East China Normal University, Shanghai 20062, China shenjh@stu.ecnu.edu.cn, {hqian, amzhou}@cs.ecnu.edu.cn, zhangwei.thu2011@gmail.com
Pseudocode Yes Algorithm 1 shows the implementation of SCDM.
Open Source Code Yes The source code of SCDM is available at Git Hub1.
Open Datasets Yes The experiments are conducted on four real-world datasets, i.e., Math1, Math2 (Liu et al. 2018), Frac Sub (Wu et al. 2015) and Neur IPS20 (Wang et al. 2020b).
Dataset Splits No The paper states: "The size of test dataset is 0.2", indicating a 20% test split. However, it does not explicitly specify the division of the remaining 80% into training and validation sets, or mention k-fold cross-validation, which would provide the full dataset split information.
Hardware Specification No The paper mentions the use of specific software frameworks like PyTorch and optimizers like Adam, but it does not provide any concrete details about the hardware used (e.g., specific GPU models, CPU types, or memory).
Software Dependencies No The paper mentions "DEAP (De Rainville et al. 2012)" and "Pytorch (Paszke et al. 2019)" as implemented software. While it cites the papers for these tools, it does not provide specific version numbers for these software components, which is required for reproducibility.
Experiment Setup Yes In the GP module implemented by DEAP (De Rainville et al. 2012), the population size V is 200, the number of generation TGP is 10, the crossover and mutation rates are 0.5 and 0.1 respectively, the initial tree depth is 5, and the selection method is tournament selection. ... In the continuous optimization module implemented by Pytorch (Paszke et al. 2019), we set the learning rate of Adam (Kingma and Ba 2015) to be 0.002, and initialize the feature parameters in the interaction function with Xavier normal initialization (Glorot and Bengio 2010).