作文机器评测中的篇章语义理解关键技术研究及应用

在国内大中小学，每年都有大量的中英文作文需要人工批改，耗时耗力，作文机器批改技术可以解决这个问题。作文的篇章语义智能评测是作文机器批改的关键环节，虽有相关研究，但是大多采取分类和回归等方法进行整体评测，而篇章语义的评测维度多样，需要深度语义分析才可解决，因此需要对作文的篇章语义进行多维度精细化评测研究，突破目前篇章语义评测精准率不高的技术瓶颈。本项目针对大中小学作文篇章语义评测的共性关键技术问题进行研究，首先突破高层语义模型、实体及关系联合抽取、语义块的自动分割等关键技术，然后研究作文知识图谱的构建、主题一致性分析、篇章结构合理性分析、语言描写能力分析、句间逻辑合理性分析等技术，进而融合上述模型和技术完成作文篇章语义评测模型的研究。研究成果将应用于“批改网”等现有的中英文作文智能评测系统，为教师减负、学生提高语言学习效率服务，为疫情、雾霾等期间的网络学习服务，为国家的教育均衡战略服务。

In domestic primary and middle schools, there are a large number of Chinese and English compositions that need to be corrected manually every year, which is time-consuming and labor-consuming.This problem can be solved by the technology of composition machine correcting. Intelligent evaluation of text semantics is the key link of composition machine evaluation. Although there are relevant researches, most of them adopt the methods of classification and regression for overall evaluation. However, the evaluation dimensions of text semantics are diverse, which can only be solved by deep semantic analysis. Therefore, it is necessary to conduct multi-dimensional and refined evaluation research on the text semantics of composition to break through the current accuracy rate of text semantic evaluation Low technology bottleneck. This project focuses on the research of common key technologies in the semantic evaluation of composition texts in primary and middle schools. Firstly, the project breaks through the key technologies such as high-level semantic model, joint extraction of entities and relationships, automatic segmentation of semantic blocks, and then studies the construction of composition knowledge graph, the assessment of theme consistency, the assessment of text structure rationality, the assessment of language description ability, and the logical rationality between sentences. Then, we integrate the above models and techniques to complete the research of the semantic evaluation model of the composition text. The research results will be applied to the existing Chinese and English composition intelligent evaluation system, such as "correction network", to serve teachers' burden reduction, students' language learning efficiency improvement, network learning during the epidemic, haze and other periods, and national education equilibrium strategy.