基于预训练模型和Transformer架构的大数据与计算机类科普书籍难度分类研究  

Research on Difficulty Classification of Big Data and Computer Popular Science Books Based on Pretrained Models and Transformer Architecture

作　　者：黄启洲 HUANG Qizhou(Unicom Digital Technology Co.,Ltd.,Beijing 100032)

机构地区：[1]联通数字科技有限公司,北京100032

出　　处：《软件》2024年第7期153-155,共3页Software

摘　　要：针对当前研究在书籍级长文本可读性评估方面的不足,本文提出了一种新颖的PTDE-CAC模型。该模型将书籍分割为固定片段,利用无监督聚类获取难度感知片段,对预训练模型进行再训练,使其学习难度知识,将长文本表示为多个不同难度级别的向量。本文构建了大数据、计算机科普教材分级数据集,实验证明PTDE-CAC模型在可读性评估中表现优异,优于传统方法和现有预训练模型。本工作为书籍级可读性评估提供了新思路,也为相关教材编写选择提供了参考。To address the inadequacy in book-level long text readability assessment,we propose a novel PTDE-CAC model.It divides books into fixed segments,obtains difficulty-aware segments via unsupervised clustering,and retrains a pre-trained model to learn difficulty knowledge,representing long texts as multiple vectors with different difficulty levels.This article construct a graded dataset of big data and computer science popular textbooks.Experiments prove PTDE-CAC outperforms traditional methods and existing pre-trained models in readability assessment.This work provides a new approach for book-level readability assessment and a reference for relevant textbook compilation and selection.

关 键 词：书籍级长文本 可读性评估 PTDE-CAC模型 难度感知预训练 多视角表示 大数据 计算机科普教材分级数据集 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]