一种作文离题检测方法

• Format: Patent
• Language: Chinese
• Published: 23.06.2020
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

本公开实施例涉及一种作文离题检测方法，其包括：对作文集进行主题模型训练，得到作文的Biterm-LDA主题模型，并对文档集进行Doc2vec模型训练，得到Doc2vec文档向量模型；将Biterm-LDA主题模型与Doc2vec文档向量模型的文本表示进行组合，得到组合特征；对作文的组合特征基于孪生网络的多层感知机进行降维和特征优化；对降维和特征优化后的话题作文，分为切题作文和离题作文，对切题作文中的一部分构建话题类中心，并根据话题类中心对切题作文中的剩余部分和离题作文分别计算，得到针对同一话题的一组阈值；根据一组阈值利用ROC曲线进行筛选，得到最佳阈值。本公开能为不同话题作文动态计算最佳阈值。 A written composition off-topic detection method, comprising: performing subject model training on a written composition set to obtain a written composition Biterm-LDA subject model, and performing Doc2vec model training on a document set to obtain a Doc2vec document vector model (S110); combining the text representation of the Doc2vec document vector model and the Biterm-LDA subject model to obtain combined features (S120); on the basis of a twin network multi-layer perceptron, performing dimensionality reduction and feature optimisation of the written composition combined features (S130); dividing topic written compositions after dimensionality reduction and feature optimisation into on-topic written composition