# Dynamic emotion recognition and learning motivation prediction in Chinese second language acquisition under cultural differences: a study based on ED-CM-MP model

## TL;DR

This paper introduces a new model for predicting learning motivation in Chinese language acquisition that adapts to cultural differences and improves accuracy and efficiency.

## Contribution

The ED-CM-MP model integrates dynamic emotion recognition, cultural adjustment, and efficient prediction for cross-cultural language learning.

## Key findings

- The model achieves high accuracy with MAE of 0.28–0.29 and F1 Score of 0.91–0.92.
- It shows improved efficiency with inference latency of 38.5–39.2 ms and FLOPs of 12.6–13.1 G.
- The model demonstrates strong cross-cultural generalization with a cultural adaptation score of 0.94–0.95.

## Abstract

To address the core issues of low accuracy, poor cultural adaptation, and insufficient efficiency in learning motivation prediction in cross-cultural Chinese second language acquisition scenarios, this paper proposes the ED-CM-MP model, which integrates dynamic sentiment recognition, cultural adjustment modeling, and lightweight temporal prediction. This model uses DistilBERT+Gated TCN to construct a dynamic sentiment module to extract temporal sentiment features, GraphSAGE to adjust for cross-cultural differences, and Temporal Fusion Transformer to achieve efficient motivation prediction. Experiments on the HSK and VIDAS cross-cultural datasets show that the model achieves the best core prediction performance: MAE of 0.28–0.29 and F1 Score of 0.91–0.92, representing a 10.2% improvement in accuracy compared to the best baseline model; inference latency as low as 38.5–39.2 ms and FLOPs of only 12.6–13.1 G, representing a 20.3% improvement in efficiency compared to MobileNetV3; and a cultural adaptation score of 0.94–0.95, representing a 21.8% improvement in cross-cultural generalization ability compared to U-Cast. Ablation experiments validated the necessity of the three modules working together; removing any module resulted in a performance decrease of 3.6%-7.2%. Stability tests showed that the model exhibited excellent robustness with performance fluctuations of ≤ 5.4% in a small sample scenario with 10% labeled noise and 2000 training samples. This research demonstrates that the ED-CM-MP model achieves a triple breakthrough in motivation prediction–accuracy, efficiency, and generalization–providing an efficient and feasible technical solution for intelligent teaching intervention in cross-cultural Chinese second language acquisition.