语篇重读的眼动特征及预测模型
文献类型：学位论文
| 作者 | 徐颖 |
| 答辩日期 | 2024-06 |
| 文献子类 | 硕士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 王利刚 |
| 关键词 | 重读效应 数字阅读 眼动技术 神经网络模型 |
| 学位名称 | 应用心理硕士 |
| 学位专业 | 应用心理 |
| 其他题名 | Eye Movement Characteristics of Discourse Rereading and Prediction Model Using Artificial Neural Networks |
| 中文摘要 | The strategy of rereading is a commonly used reading comprehension strategy, which can effectively help readers read more quickly, memorize details better, deepen their comprehension of the text, and improve reading efficiency. However, the cognitive mechanisms underlying the differences between the first reading and rereading are still unclear. Therefore, it is vitally important to explore the cognitive characteristics and evaluation methods of the rereading effect. This study consists of two sub-studies. Study 1 employed an eye-tracking experimental approach with 103 college students as participants, who were divided into two groups based on their reading proficiency (high and low). These participants all speak English as a second language. The participants were asked to read and evaluate the discourse materials twice, and their eye movement characteristics and gaze distribution during the first reading and rereading processes were recorded in real time. In Study 2, 20 significantly different eye movement indicators and 7 time series indicators based on saccades were extracted from Study 1 as input valuables. The objective was to investigate the feasibility of constructing an effective neural network eye movement model for classifying the first reading and rereading. The results of Study 1 demonstrated that eye movement indicators during discourse reading were applicable for classifying the first reading and rereading. During the rereading process, participants exhibited shorter total reading time (TRT), average fixation duration (AFD), average regression distance across lines (RDAL) and the specific interest area (IA) indicators including total fixation time (IA_TFD), average first fixation duration (IA_FFD), average first-pass reading time (IA_FRT), average second fixation duration (IA_SFD), and average second-pass reading time (IA_SRT). Participants also exhibited longer average regression size (ARS) and average regression distance within line (RDWL). Additionally, participants exhibited fewer total number of fixations (TNF), total regression counts (TRC), average regression counts within line (RCWL), average regression counts across lines (RCAL) and the specific interest area (IA) indicators including total number of fixations (IA_TNF), average regression out counts (IA_ROC), average first fixation counts (IA_FFC), and average second fixation counts (IA_SFC). Pupil size, proportion of fixation duration, and regression duration within line and across lines were not found to be useful in identifying rereading. In Study 2, the rereading neural network classification model based on overall eye movement indicators achieved high accuracy (0.769), precision (0.774), recall (0.788), and F1-score (0.781). Among these indicators, local interest area eye movement indicators were found to be more important than global eye movement indicators in identifying rereading. Furthermore, the neural network classification model based on time series indicators of saccades also demonstrated good classification performance (accuracy, precision, recall, and F1-score were all 0.733). Among these indicators, the counts of saccade differential values had the most significant impact on the overall predictive performance of the model, while the mean and standard deviation of saccade differential values had the smallest impact. Generally speaking, the eye movement characteristics and fixation distribution of the first reading and rereading are quite different. Compared with the first reading, the total reading time, fixation time, regression distance, regression counts and local eye movement behaviors are all reduced. In addition, the neural network model with global eye movement indicators and saccade time series indicators as inputs can effectively classify first reading and rereading, and can be used to evaluate readers' text familiarity. |
| 英文摘要 | 重读是一种常用的阅读理解策略，能有效帮助学生加深对文本的理解，提高阅读效率。但在首次阅读和重读时其认知机制有何不同尚不明确。因此有必要探究重读效应的认知特点和评估方法。本研究包括两个子研究。研究一采用眼动阅读实验的方法，选取 103 名以英语为二语的大学生为被试，将其划分为高、低阅读水平两组，对英文语篇材料进行两次阅读和测评，并采用眼动追踪技术实时记录了读者在首次阅读和重读过程中的眼动特征和注视分布情况，探究重读对眼动行为的影响；研究二分别从研究一中提取 20 个差异显著的眼动指标和七个眼跳序列指标作为输入，探究是否可以构建有效的神经网络眼动模型对首次阅读和重 读进行分类。 研究一的结果表明，语篇阅读过程中的眼动指标适用于分类首次阅读和重读。 在重读过程中，被试的总阅读时间（TRT）、平均注视时间（AFD）、跨行平均 回视距离（RDAL）和局部兴趣区内的总注视时间（IA_TFD）、平均首次注视 时间（IA_FFD）、平均第一遍阅读时间（IA_FRT）、平均第二次注视时间（IA_SFD）、平均第二遍阅读时间（IA_SRT）较短；平均回视距离（ARS）和行内平均回视 距离距离（RDWL）较长；总注视次数（TNF）、总回视次数（TRC）、行内回 视次数（RCWL）、跨行回视次数（RCAL）以及局部兴趣区内的总注视次数 （IA_TNF）、平均回视出次数（IA_ROC）、平均第一遍注视次数（IA_FFC）和平均第二遍注视次数（IA_SFC）较少。瞳孔大小、注视时间的比例、行内和跨行的回视时间对识别重读没有帮助。研究二中，基于总体眼动指标的重读神经 网络分类模型获得了较高的准确率(accuracy = 0.769，precision = 0.774，recall = 0.788，F1-score = 0.781)。其中，局部眼动指标在识别重读过程中比全局眼动指 标更重要。另外，基于眼跳序列指标的神经网络分类模型也表现出较好的分类效果（accuracy、precision、recall 和 F1-score 均为 0.733）。其中，眼跳差值次数对 模型的整体预测性能影响最大，均值和标准差对模型的影响最小。 总的来说，首次阅读和重读的眼动特征以及注视分布情况差异较大。与首次阅读相比，重读的总阅读时间、注视时间、回视距离、回视次数和兴趣区的局部眼动行为都有所减少。此外，以总体眼动指标以及眼跳序列指标为输入的神经网 络模型均可以有效地分类首次阅读和重读，可以应用于评估读者的文本熟悉度。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/48156]  |
| 专题 | 心理研究所_健康与遗传心理学研究室 |
| 推荐引用方式 GB/T 7714 | 徐颖. 语篇重读的眼动特征及预测模型[D]. 中国科学院心理研究所. 中国科学院大学. 2024. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。