Short Answer Automatic Scoring Based on Multi-model Dynamic Collaboration∗
Yuze Wu, Xiaopeng Cao, Xiang Tian · 2021 4th International Conference on Artificial Intelligence and Pattern Recognition · 2021
Short answer automatic scoring is a research hotspot of text semantic matching. Education text contains the multidisciplinary characteristics. Each discipline can be subdivided into a small field. Researchers use a large-scale and hyper-parametric deep learning model to deal with these text. This method results in difficult reproduction and high training costs. In this paper, we propose a Multi-model Dynamically Cooperation Semantic Matching (MMDCSM) algorithm which combines multiple models through rule collaboration. We use LCQMA dataset and Xiyou ASAG dataset to experiment MMDCSM algorithm. Xiyou ASAG dataset is created by Xi'an University of Posts and Telecommunication. The results show that MMDCSM algorithm can effectively improve the accuracy and reduce the complexity of related training in the short answer automatic scoring task.