Research and Application of Chinese Composition Automatic Scoring System Based on Reinforcement Learning

With the rapid development of educational informatization, traditional essay scoring methods are inefficient and subjective, making it difficult to adapt to the needs of large-scale evaluation. This paper aims to study an automatic scoring system for Chinese essays based on reinforcement learning. First, a large amount of annotated Chinese essay data is collected to construct a training set (containing 10,000 scored essays) and a test set (containing 2,000 essays). Secondly, essay features, including vocabulary richness, grammatical correctness, content coherence, etc., are extracted to form feature vectors. Then, a reinforcement learning model is designed, using deep Q-learning (DQN). The feature vector is used as input and the scoring result is used as action. The reward signal is calculated by comparing with the manual scoring to guide model learning. Finally, through multiple rounds of training, the model parameters, such as learning rate and discount factor, are adjusted to optimize the scoring strategy. The proposed scoring system has a maximum accuracy of 96.4% on the test. The automatic scoring system based on reinforcement learning can not only effectively reduce the scoring time, but also provide more objective and consistent scoring results.