结合学科情感分析与依存关系的相似度评分

通过对语文古诗文阅读类主观题的分析,提出了结合学科情感分析与依存关系的相似度评分算法,并将其应用于高中语文古诗文阅读类主观题的评分中。首先,以中文维基百科语料为基础,增加了与评分相关的古诗文语料81 927条,通过文本向量化算法Word2vec进行词向量训练,完成了对古诗文语料库的构建;基于学科评分特性建立了对应的古诗文过滤词表,提出了基于词性的关键词提取及词向量的相似度计算方法;之后,针对情感分析模型对古诗文语句分析不准确的问题,结合同义词词林,建立了古诗文情感词库;并构建了学科情感分析模型,实现了基于学科情感分析的相似度计算方法;最后,通过关键词、学科情感分析以及依存句法分析,从多个维度计算学生答案与标准答案文本之间的加权语义相似度。并将构建的古诗文语料库、古诗文情感词库和学科情感分析模型,用于相似度综合评分算法,以此实现了结合学科情感分析与依存关系的相似度评分算法。实验表明,该算法的平均评分准确率达到了89.42%。

Similarity Score Combining Subject Sentiment Analysis and Dependency Relationship

Through the analysis of the subjective questions of ancient Chinese poetry reading, a similarity scoring algorithm combining subject sentiment analysis and dependency relationship is proposed and applied to the scoring of the subjective questions of ancient Chinese poetry reading in high school. First of all, based on the Chinese Wikipedia corpus, 81 927 pieces of ancient poetry corpus related to scoring are added, and word vector training is carried out through the text vectorization algorithm Word2 vec, and the construction of the ancient poetry corpus is completed. Based on the subject scoring characteristics, the corresponding ancient poetry filter vocabulary is established, and a method of keyword extraction based on part of speech and the similarity calculation method of word vectors is proposed. After that, in order to solve the problem of inaccurate analysis of ancient poetry sentences by the sentiment analysis model, combined with the synonym word forest, the ancient poetry sentiment word library is established, and a subject sentiment analysis model is constructed to realize the similarity calculation method based on subject sentiment analysis. Finally, the weighted semantic similarity between students’ answers and standard answer texts is calculated from multiple dimensions through keywords, subject sentiment analysis and dependency analysis. And the constructed ancient poetry corpus, ancient poetry sentiment vocabulary and subject sentiment analysis model are used in the similarity comprehensive scoring algorithm, so as to realize the similarity scoring algorithm combining subject sentiment analysis and dependency relationship. Experiments show that the average scoring accuracy of the algorithm reaches 89.42%.