研究目的
To implement a methodology for analyzing and classifying large volumes of eye tracking records using data mining to study visual attention in people with different levels of orthographic knowledge during spelling error and character detection tasks.
研究成果
Data mining techniques effectively revealed differences in eye movements based on orthographic knowledge levels, with spelling errors influencing visual attention in high-knowledge participants. However, no clear patterns were found in clustering related to behavioral outcomes, suggesting that eye movement strategies may not directly correlate with task performance or knowledge levels.
研究不足
The study was limited to Spanish-speaking participants and specific tasks, potentially not generalizable to other languages or contexts. The sample size was reduced from 45 to 36 due to data integrity issues, which may affect statistical power. The eye tracker's sampling rate and accuracy could impose technical constraints.
1：Experimental Design and Method Selection:
The study used data mining techniques including decision trees and k-means clustering to analyze eye movement data from an eye-tracking device. The methodology involved creating secondary variables, searching for response patterns, and developing new models based on distances from gaze positions to word centers and spelling errors.
2：Sample Selection and Data Sources:
Forty-five Spanish-speaking subjects were selected and performed two tasks: Explicit Error Detection Task (EEDT) and Implicit Error Detection Task (IEDT). Data included behavioral records and eye movements recorded with an SMI RED-m eye tracker.
3：List of Experimental Equipment and Materials:
SMI RED-m eye tracker (SensoMotoric Instruments), software iView, Experiment Center 3.4, and BeGaze 3.4 for data recording and analysis. WEKA 3 software for data mining algorithms.
4：4, and BeGaze 4 for data recording and analysis. WEKA 3 software for data mining algorithms.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Participants were exposed to words with and without misspellings for 1500 milliseconds. Eye movements were recorded at 60-120 Hz. Data preprocessing included creating new variables, integrity checks, and applying classification and clustering algorithms.
5：Data Analysis Methods:
Used decision tree algorithms (J48 and Random Forest in WEKA) for variable classification and k-means for unsupervised clustering. Analyzed behavioral results, reaction times, and eye movement patterns.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
