Complex DNN architectures offer little advantage over simple linear ridge-regression baselines for trait scoring, implying nonlinearities and interactions were of minor importance¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Comparisons to a feature-based linear regression baseline only partly revealed advantages for the DNN target models, and the T-tests were not significant. → Lohmann 2025

Evidence¶

Lohmann 2025¶

Lohmann, J. F., Junge, F., Möller, J., Fleckenstein, J., Trüb, R., Keller, S., Jansen, T., & Horbach, A. (2025). Neural Networks or Linguistic Features? - Comparing Different Machine-Learning Approaches for Automated Assessment of Text Quality Traits Among L1- and L2-Learners' Argumentative Essays. International Journal of Artificial Intelligence in Education, 35, 1178–1217. https://doi.org/10.1007/s40593-024-00426-w

q2 · i? · design · r2

Baseline comparisons in the main experiments used ridge regression with the 220-feature set; the feature DNN even fell short in four of six MEWS traits (average ΔQWK −0.02), while ASAP advantages were small (0.02 and 0.01).

"comparisons to the feature-based linear regression baseline only partly revealed advantages for the target models, and T-tests were not significant ( pfeatures = 0.818 , pembeddings = 0.465 )."

Discussion¶

Related Claims¶

• ANN analysis identifies AI learning self-efficacy and perceived usefulness as the strongest and most stable predictors of AI learning anxiety, with only modest advantage over linear benchmarks — related
• Quadratic and cubic trends significantly improved the linear model of teacher and student behavior, showing the importance of studying nonlinearity — related
• For trait-specific automated essay scoring trained on a single resource, feature-based DNNs slightly outperform embedding-based DNNs, winning 11 of 16 trait-prompt comparisons — related
• Ablation tests identify unique trait-relevant information in specific feature groups: readability and syntactic complexity for content, cohesion and readability for organization, and occurrence, length, or error features for language — reports the opposite
• Addition tests show trait-specific feature groups that most consistently enhance embedding-based models: morphological complexity for content, length for organization, and lexical sophistication, error, and occurrence features for language — reports the opposite