文献类型：期刊文献

英文题名：Research on Personalized Learning Strategies of College English Based on Reinforcement Learning Algorithm in the Era of Internet+Intercultural Education

作者：Ni, Huaying[1]; Shen, Nitianchou[2]

机构：[1] Shangyu College, Shaoxing University, Zhejiang, Shaoxing, 312300, China; [2] Wenzhou Business College, Zhejiang, Wenzhou, 325035, China

年份：2025

卷号：46

期号：3

起止页码：1076

外文期刊名：International Journal for Housing Science and Its Applications

收录：EI(收录号：20253519046180)、Scopus(收录号：2-s2.0-105014025814)

语种：英文

外文关键词：Convolution - Convolutional neural networks - E-learning - Learning algorithms - Learning systems - Mean square error - Prediction models - Reinforcement learning - Semantics - Statistical tests

外文摘要：Personalized learning is a strategy that recommends the best learning strategies (including learning resources, test questions, etc.) based on the individual learner's situation, so that the learner can obtain the optimal development. This paper takes the automatic prediction and recommendation of test question difficulty as the research object, and proposes a convolutional neural network model based on semantic attention mechanism for college English test questions. The model is based on the practice characterization method of semantic comprehension to extract semantic features and quantify the semantic dependence degree of English reading test questions, so as to assess the quality of reading test questions. At the same time, the automatic prediction model of test question difficulty is constructed by taking into account the content, difficulty and objectives of multiple test question types in English learning. The model is based on convolutional neural network and deep attention network algorithms to realize automatic prediction of English test difficulty. And through reinforcement learning settings, personalized rewards are set to guide the recommendation, and the test question recommendation model is constructed by combining the difficulty prediction of test questions. The model always shows optimal results in predicting the difficulty of test questions on different datasets, and the average absolute error is lower than 0.33 and the root mean square error is lower than 0.38, which demonstrates the high reliability of personalized recommendation of test questions. ? 2025 International Association for Housing Science. All rights reserved.