Prediction Model of Students' Learning Motivation Based on Combinatorial Optimization Algorithm

Weifeng Deng, Lin WangInternational Journal of Emerging Technologies in Learning (iJET) 2023 · DOI

Abstract

The strength of college students' learning enthusiasm directly affects their learning effect. Studying and predicting students' learning enthusiasm has positive significance for improving college students themselves and college education. Existing methods have not involved how to select and extract features related to students' learning enthusiasm, so it is difficult to meet the needs of learning enthusiasm prediction. Therefore, this article takes English learning as an example to conduct a research on the prediction model of students' learning enthusiasm based on combinatorial optimization algorithm. Taking English learning as an example, this article expounds in detail the main manifestations of students' learning enthusiasm, and predicts the intention of students to participate in learning behaviors based on the enumerated behaviors, so as to judge students' learning enthusiasm, and gives the construction method of the prediction model. This study constructs a network model for the evaluation of students' learning autonomy, and finally outputs the grade results of students' learning autonomy evaluation. Based on the obtained predictions of students' participation in learning behavior intentions and the evaluation results of students' learning autonomy, a combined model is established through the weighting of the inverted error method to predict the regression of students' learning enthusiasm. Experimental results verify the effectiveness of the constructed single model and combined model.

Introduction

Student motivation—or "learning enthusiasm"—is a primary driver of educational success, especially in higher education where students have a high degree of autonomy. Educators and automated systems benefit greatly from being able to predict this enthusiasm to provide timely interventions. However, existing educational data mining and machine learning methods struggle to adapt across different subjects and student demographics, often leading to over-fitting or under-fitting. More importantly, previous methods lack a principled way to select and extract the specific behavioral features that genuinely reflect enthusiasm.

Conceptual Framework and Feature Selection

The authors bridge the gap in feature selection by explicitly capturing both the temporal sequence of student behaviors and their underlying, personalized intentions. As shown in Figure 1, the analysis is categorized into three main pillars: basic personal information, behavioral intention prediction, and learning autonomy evaluation. These features provide the necessary input for a more nuanced understanding of student engagement.

Figure 1: Framework for analyzing changes in college students' learning enthusiasm, detailing the specific features and behaviors tracked.

Proposed Methodology

The proposed framework consists of three main components designed to handle the complexity of educational data:

1. Intention Prediction Model

This model processes a student's historical learning behaviors. It uses a time-aware LSTM to handle the timestamps of interactions. The output is fed into a multi-intent self-attention module. Crucially, a directional mask matrix is applied to this attention mechanism to ensure the model only looks at past behaviors, preserving the chronological reality of learning. Finally, a personalized attention network injects student metadata (age, gender, major) to account for individual preferences.

Figure 3: Detailed architecture of the intention prediction module, integrating time-aware LSTM units with multi-intent self-attention and personalized networks.

2. Autonomy Evaluation Model

This separate network grades a student's learning autonomy using a spatial self-attention module and a polarity detection branch. It is trained using a multi-task loss function to simultaneously predict coarse-grained and fine-grained autonomy categories, providing a multi-dimensional view of student independence.

3. Combinatorial Optimization (Inverted Error Method)

To generate the final "learning enthusiasm" score, predictions from the Intention and Autonomy models are combined. The authors use an inverted error weighting method: each model is assigned a weight based on the reciprocal of its prediction error (specifically using RMSE). This ensures that the more accurate model exerts more influence on the final prediction, resulting in a more robust ensemble output.

Key Results and Insights

  • Unidirectional Attention: Adding a directional mask matrix prevents "future" behaviors from influencing the representation of "past" behaviors, which is vital for chronological educational data.
  • Personalization: Incorporating static student embeddings (like major and age) alongside dynamic behavioral sequences helps the model understand the context behind different students' intentions.
  • Model Performance: The optimized intention prediction model achieved an $R^2$ of 0.965. The final combined model, utilizing RMSE-based inverted error weighting, achieved an $R^2$ of 0.956, significantly outperforming a traditional BP neural network baseline ($R^2$ of 0.285).
  • Hyperparameter Optimization: The best performance was achieved with an LSTM hidden layer dimension of 64 and 8 attention heads. Too few heads failed to capture complex intentions, while too many introduced noise.

Conclusion

This research demonstrates that combining temporal behavioral modeling with autonomy evaluation through a weighted ensemble significantly improves the prediction of student learning enthusiasm. While the study focuses on English learning, the methodology provides a blueprint for more sophisticated educational intervention systems. Future work should focus on testing this model across diverse academic subjects and against a broader range of state-of-the-art baselines to further validate its generalizability.

Summary updated

Abstract

The strength of college students' learning enthusiasm directly affects their learning effect. Studying and predicting students' learning enthusiasm has positive significance for improving college students themselves and college education. Existing methods have not involved how to select and extract features related to students' learning enthusiasm, so it is difficult to meet the needs of learning enthusiasm prediction. Therefore, this article takes English learning as an example to conduct a research on the prediction model of students' learning enthusiasm based on combinatorial optimization algorithm. Taking English learning as an example, this article expounds in detail the main manifestations of students' learning enthusiasm, and predicts the intention of students to participate in learning behaviors based on the enumerated behaviors, so as to judge students' learning enthusiasm, and gives the construction method of the prediction model. This study constructs a network model for the evaluation of students' learning autonomy, and finally outputs the grade results of students' learning autonomy evaluation. Based on the obtained predictions of students' participation in learning behavior intentions and the evaluation results of students' learning autonomy, a combined model is established through the weighting of the inverted error method to predict the regression of students' learning enthusiasm. Experimental results verify the effectiveness of the constructed single model and combined model.
BibTeX
@article{deng2023prediction,
  author = {Weifeng Deng and Lin Wang},
  title = {{Prediction Model of Students' Learning Motivation Based on Combinatorial Optimization Algorithm}},
  journal = {International Journal of Emerging Technologies in Learning (iJET)},
  year = {2023},
  doi = {10.3991/ijet.v18i09.40229},
}
Figure