基于SVM的眼动轨迹解读思维状态的研究
|
摘要
眼动轨迹是眼睛注视事物过程中注视点随时间变化的序列,它动态反映人在阅读、驾驶、运动等过程中的眼部活动情况,眼动轨迹含有非常丰富的信息。
目前,从人们的眼动轨迹来解读人的思维状态已成为应用心理学的研究热点。然而,在眼动研究的众多领域中,如:阅读的眼动研究;图画观看、视觉搜索和模式识别的眼动研究;眼动的交通心理学研究;眼动在航空心理学研究中的应用、眼动在体育心理学研究中的应用等。眼动研究所采用的实验方法是:首先对被试的眼动参数(如注视时间、眼跳时间、回视时间、眼跳潜伏期、追随运动时间等)进行统计,再对统计结果进行统计学分析(如多因素方差分析),最后对统计学的结果进行分析得出实验结论。这样的研究方法都是对眼动轨迹的静态指标的分析,缺少对眼动轨迹整个动态指标的分析。
为此,本研究以眼动轨迹动态指标分析方法为研究重点,结合先进的数据挖掘技术对眼动轨迹进行分析。本文以四方趣题为研究材料,通过Tobii眼动仪记录被试解题时的眼动轨迹,然后对记录的眼动轨迹数据采用综合眼动指标加权的方法进行预处理,并采用机器学习中的SVM方法对预处理后的眼动轨迹进行分析,从而通过眼动轨迹来解读人的思维状态。
本文的研究方法可被推广到其他领域,如,分析驾驶员的眼动轨迹,合理设计交通管理办法,避免交通事故;分析阅读时眼动轨迹设计合理人机界面等。
本文主要做了以下几方面的工作:
1、为了实现眼动轨迹解读人的思维状态,本文采用了四方趣题做为研究材料,通过Tobii眼动仪来记录被试解答四方趣题过程中的眼动轨迹,从而从记录的眼动轨迹中来解读人们解答四方趣题时的思维状态。
2、数据预处理是数据挖掘中最关键的一步,它关系到挖掘的质量。传统的眼动研究对数据按照注视时间、回视时间等单个眼动指标进行统计分析,不能完整反映眼动信息。本文采用的预处理方法是:对四方趣题中的各个兴趣区的注视持续时间和回视等眼动综合指标进行加权求值,从而全面地反映眼动轨迹所包含的信息。
3、支持向量机(Support Vector Machine-SVM)作为新出现的机器学习方法,以其良好的分类性能受到广泛关注,取得了丰硕的研究成果。本文采用SVM对数据预处理后的眼动轨迹数据进行二分类,然后根据分类的结果来解读解题的思维状态。
4、本文中的四方趣题解题策略分为七种,而传统的SVM是一种二分类的方法,为此,本文采用二叉树多分类SVM算法与聚类相结合的思想,构建一个多分类的二叉决策树层次结构,从而实现解题策略的多分类。
Eye movement trajectory is the sequence of gaze points changing along with the time when eye is gazing the things. It reflects dynamically people’s eye movements when they are reading, driving, doing sports, and so on. So it contains abundant information. At present, to interpret people’s mental state from their eye movement
trajectory has become a research hotspot in applied psychology. However, in many fields about eye movement research, such as, eye movement research on reading, eye movement research on watching pictures, visual search and pattern recognition, eye movement research on traffic psychology, the application of eye movement in the study of aviation psychology, the application of eye movement in sports psychology research, and so on. The experimental method applied in the eye movement research is that, firstly, count eye indicators (such as, fixation time, eye dancing time, regression time, eye twitching latent period, follow exercise time, etc), secondly, analyze statistically the statistical results (e.g. multi-factor variance analysis), finally, the experimental results are obtained according to statistical results. Such research method only analyzes the static indicators of eye movement trajectory, and is lack of analysis of uniform dynamic indicators of eye movement trajectory.
Therefore, this study surrounds closely the purpose of proposing the method of analyzing dynamic indicators of eye movement trajectory, and combines with advanced data mining technology to analyze eye movement trajectory. This paper uses 4*4 Sudoku as research material. Tobii eye tracking system records the eye movements when subjects are solving the 4*4 Sudoku. Then preprocess the recorded eye movement trajectory data with the method of weighting the integrated eye indicators, and analyze the preprocessed eye movement trajectory data with SVM from machine learning. In the end, interpret people’s mental state from eye movement trajectory.
The research methods in this paper can be applied in other fields, such as, analyze drivers’eye movement trajectory, and design rational traffic management rules to prevent traffic accidents; analyze people’s eye movement trajectory when they are reading to design reasonable man-machine interface, and so on.
In this paper, the main works are as follows:
1、In order to interpret people’s mental state from eye movement trajectory, this paper uses 4*4 Sudoku as research material. Tobii eye tracking system records the subjects’eye movement trajectory when they are answering 4*4 Sudoku. And then interpret people’s mental state when they are solving 4*4 Sudoku from the recorded eye movement trajectory.
2、Data preprocessing is the most critical step in data mining, and it influences the quality of data mining. The traditional eye movement research analyzes statistically the single eye indicator (such as, fixation time, regression time, and so on), but can not completely reflect the eye movement information. In this paper, the preprocessing method is that, weighting the eye composite indicators of every area of interest in 4*4 Sudoku, and fully reflecting the eye movement trajectory.
3、Support vector machine(SVM)as a new emerging machine learning method, with its good classification performance, attracted widespread attention, has achieved fruitful research results. This paper uses SVM to classify the eye data preprocessed, and then interprets the mental state according to the classification results.
4、The 4*4 Sudoku contains seven problem solving strategies, whereas the traditional SVM is two-classification method. So this paper combines binary tree multi-classification SVM algorithm with clustering, and establishes multi-classification binary decision tree hierarchy, and achieves the multi-classification of problem solving strategies.
目前,从人们的眼动轨迹来解读人的思维状态已成为应用心理学的研究热点。然而,在眼动研究的众多领域中,如:阅读的眼动研究;图画观看、视觉搜索和模式识别的眼动研究;眼动的交通心理学研究;眼动在航空心理学研究中的应用、眼动在体育心理学研究中的应用等。眼动研究所采用的实验方法是:首先对被试的眼动参数(如注视时间、眼跳时间、回视时间、眼跳潜伏期、追随运动时间等)进行统计,再对统计结果进行统计学分析(如多因素方差分析),最后对统计学的结果进行分析得出实验结论。这样的研究方法都是对眼动轨迹的静态指标的分析,缺少对眼动轨迹整个动态指标的分析。
为此,本研究以眼动轨迹动态指标分析方法为研究重点,结合先进的数据挖掘技术对眼动轨迹进行分析。本文以四方趣题为研究材料,通过Tobii眼动仪记录被试解题时的眼动轨迹,然后对记录的眼动轨迹数据采用综合眼动指标加权的方法进行预处理,并采用机器学习中的SVM方法对预处理后的眼动轨迹进行分析,从而通过眼动轨迹来解读人的思维状态。
本文的研究方法可被推广到其他领域,如,分析驾驶员的眼动轨迹,合理设计交通管理办法,避免交通事故;分析阅读时眼动轨迹设计合理人机界面等。
本文主要做了以下几方面的工作:
1、为了实现眼动轨迹解读人的思维状态,本文采用了四方趣题做为研究材料,通过Tobii眼动仪来记录被试解答四方趣题过程中的眼动轨迹,从而从记录的眼动轨迹中来解读人们解答四方趣题时的思维状态。
2、数据预处理是数据挖掘中最关键的一步,它关系到挖掘的质量。传统的眼动研究对数据按照注视时间、回视时间等单个眼动指标进行统计分析,不能完整反映眼动信息。本文采用的预处理方法是:对四方趣题中的各个兴趣区的注视持续时间和回视等眼动综合指标进行加权求值,从而全面地反映眼动轨迹所包含的信息。
3、支持向量机(Support Vector Machine-SVM)作为新出现的机器学习方法,以其良好的分类性能受到广泛关注,取得了丰硕的研究成果。本文采用SVM对数据预处理后的眼动轨迹数据进行二分类,然后根据分类的结果来解读解题的思维状态。
4、本文中的四方趣题解题策略分为七种,而传统的SVM是一种二分类的方法,为此,本文采用二叉树多分类SVM算法与聚类相结合的思想,构建一个多分类的二叉决策树层次结构,从而实现解题策略的多分类。
Eye movement trajectory is the sequence of gaze points changing along with the time when eye is gazing the things. It reflects dynamically people’s eye movements when they are reading, driving, doing sports, and so on. So it contains abundant information. At present, to interpret people’s mental state from their eye movement
trajectory has become a research hotspot in applied psychology. However, in many fields about eye movement research, such as, eye movement research on reading, eye movement research on watching pictures, visual search and pattern recognition, eye movement research on traffic psychology, the application of eye movement in the study of aviation psychology, the application of eye movement in sports psychology research, and so on. The experimental method applied in the eye movement research is that, firstly, count eye indicators (such as, fixation time, eye dancing time, regression time, eye twitching latent period, follow exercise time, etc), secondly, analyze statistically the statistical results (e.g. multi-factor variance analysis), finally, the experimental results are obtained according to statistical results. Such research method only analyzes the static indicators of eye movement trajectory, and is lack of analysis of uniform dynamic indicators of eye movement trajectory.
Therefore, this study surrounds closely the purpose of proposing the method of analyzing dynamic indicators of eye movement trajectory, and combines with advanced data mining technology to analyze eye movement trajectory. This paper uses 4*4 Sudoku as research material. Tobii eye tracking system records the eye movements when subjects are solving the 4*4 Sudoku. Then preprocess the recorded eye movement trajectory data with the method of weighting the integrated eye indicators, and analyze the preprocessed eye movement trajectory data with SVM from machine learning. In the end, interpret people’s mental state from eye movement trajectory.
The research methods in this paper can be applied in other fields, such as, analyze drivers’eye movement trajectory, and design rational traffic management rules to prevent traffic accidents; analyze people’s eye movement trajectory when they are reading to design reasonable man-machine interface, and so on.
In this paper, the main works are as follows:
1、In order to interpret people’s mental state from eye movement trajectory, this paper uses 4*4 Sudoku as research material. Tobii eye tracking system records the subjects’eye movement trajectory when they are answering 4*4 Sudoku. And then interpret people’s mental state when they are solving 4*4 Sudoku from the recorded eye movement trajectory.
2、Data preprocessing is the most critical step in data mining, and it influences the quality of data mining. The traditional eye movement research analyzes statistically the single eye indicator (such as, fixation time, regression time, and so on), but can not completely reflect the eye movement information. In this paper, the preprocessing method is that, weighting the eye composite indicators of every area of interest in 4*4 Sudoku, and fully reflecting the eye movement trajectory.
3、Support vector machine(SVM)as a new emerging machine learning method, with its good classification performance, attracted widespread attention, has achieved fruitful research results. This paper uses SVM to classify the eye data preprocessed, and then interprets the mental state according to the classification results.
4、The 4*4 Sudoku contains seven problem solving strategies, whereas the traditional SVM is two-classification method. So this paper combines binary tree multi-classification SVM algorithm with clustering, and establishes multi-classification binary decision tree hierarchy, and achieves the multi-classification of problem solving strategies.
引文
[1]张金桥.西方关于阅读的眼动研究[J].暨南大学华文学院学报,2003,(4):73-78,
[2]阎国利,白学军.阅读过程中眼动特征的发展研究[J],心理学探新,2001,21(4):25-28.
[3]沃建中,李琪,田宏杰.不同推理水平儿童在图形推理任务中的眼动研究[J].心理发展与教育,2006,(3):6-10.
[4]阎国利.眼动分析法在广告心理学研究中的应用[J].心理学动态,1999,7(4):50-53.
[5]廖卫华,严由伟,徐子清.视线跟踪技术在人机交互中的研究[J].长春师范学院学报(自然科学版),2005,24(2):130-132.
[6]刘伟,袁修干,庄达民等.眼动测量系统及其在工效学中的应用[J].航空学报,2001,22(5):385-389.
[7]阎国利,白学军.眼动记录法在国外运动心理学研究中的应用[J].心理学动态,1997,5(2):44-48.
[8]韩玉昌.观察不同形状和颜色时眼运动的顺序性[J].心理科学,1997,20(1):40-43.
[9]阎国利,田宏杰,张仙峰.汽车驾驶行为的眼动研究[J].心理科学,2005,28(5):1211-1212.
[10]陈凯,朱钰.机器学习及其相关算法综述[J].统计与信息论坛,2007,22(5):105-112.
[11]刘琴.机器学习[J].武钢职工大学学报.2001,(6):41-44.
[12] Breiman L.Random forests[J].Machine Learning,2001,45(1):5-32.
[13]漆书青,戴海琦,丁树良.现代教育与心理测量学原理[M].南昌:江西教育出版社,1998.
[14] Cortes,C.,Vapnik,V.M.Support Vector Networks[J].Machine Learning.1995,20:273-297.
[15] Dietterich,T.G.An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees:Bagging,Boosting,and Randomization [J].Machine Learning,2000,40:139-157.
[16] Breiman,L.,Bagging Predictors[J].Machine Learning,1996,24(2):123-140.
[17] Agrawal R,Imielinski T,Swami A.Mining association rules between sets of items in large database[A][G].Proc of ACM SIGMOD Conf on Management of Data[C]. Washington,1993:207-216.
[18]刘星沙,谭利球,等.关联规则挖掘算法及其应用[J].计算机工程与科学,2007,29(1):83-86.
[19] P.Dempster,N.M.Laird,A and D.B.Rubin.Maximum Likelihood From Incomplete Data Via the EM Algorithm[J].RoyalStat.Soc.1997,39(1):1-38.
[20] Rosenblatt F . The perceptron : A perceiving and recognizing automaton(Technical report 85-4601-1)[R]. Ithaca NY:Cornell A eronautical Laboratory,1957.
[21] Vapnik V N.Estination of Dependencies Based on Empirical Data.Berlin:Spring-Verlag,1982.
[22] Vapnik V N.The Nature of Statistical Learning Theory,NY:Spring-Verlag,1995.
[23] Cherkassky V,Mulier F.Learning from Data:Concept,Theory and Methods.NY:John Viley&Sons,1997.
[24] Vapink V,Levin E,Le Cun Y.Measuring the VC-dimension of a learning machine.Neural Computation,1994,6:851-876.
[25] Burges C J C.A tutorial on support vector machines for pattern recognition.Data Mining and Knowledge Discovery,1998 2(2).
[26] V Vapnik.The Nature of Statistical Learning Theory[M].New York::Spring Verlag,1995.
[27] Buswell. Growth curves obtained from an objective study of reading habits[J].Journal of Educational Psychology,Volume 13,Issue 7,October 1922, Pages 446-447.
[28] David A. Balota,Alexander Pollatsek, Keith Rayner.The interaction of contextual constraints and parafoveal visual information in reading[J].Cognitive Psychology,Volume 17,Issue 3,July 1985,Pages 364-390.
[29]石金富,曹晓华,王钢,瞿彬.网页布局对视觉搜索影响的眼动研究[J].人类工效学,2008,14(4):1-3.
[30] Chapman P R, Undrwood G. Visual Search of Dynamic Scenes:Event Types and the Role of Experience in Viewing Driving Situations[J].In Underwood ( Eds).Eye Guidance in Reading and Scene Perception,England:Elsevier,1998:369– 394.
[31]张尧庭.广义相关系数及其应用[J].应用数学学报,1978,1(4):312-320.
[32]刘冰.多类SVM分类算法的研究和改进[J].电脑知识与技术,2007,(06):1590-1593.
[33]李昆仑.一种基于有向无环图的多类SVM分类器[J].模式识别与人工智能,2003,16(2):164-168.
[34]曾凡仔,裘正定.一种基于可行域解析中心的多类分类算法[J].复旦学报(自然科学版),2004,43(5):773-776.
[35]唐发明,王仲东,陈绵云.一种新的二叉树多类支持向量机算法[J].计算机工程与应用,2005,(07):24-26.
[36]唐发明,王仲东,陈绵云.支持向量机多类分类算法研究[J].控制与决策,2005,20(07):746-749.
[37]刘至刚等.支撑向量机在多类分类问题中的推广[J].计算机工程与应用.2004,(7):10-13.
[38] CRAMMER K,SINGER Y.On the learn ability and design of output codes for multi-class problems [C].Proceedings of the Thirteenth Annual Conference on Computational Learning Theory,San Francisco,Morgan Kaufman,2000:35-46.
[39] HSU C-W,Lin C-J.A comparison of methods for multi-class support vector machines[J ].IEEE Transaction on Neural Network ,2002,13 (2) :415– 425.
[40]边肇祺,张学工.模式识别[M].第2版.北京:清华大学出版社.2000.
[41] SCHLOPFB,MIKA S, BURGES C,et al. Input space versus feature space in kernel-based methods[J]. IEEE Transaction Neural Networks,1999,10(5):1000-1017.
[42] GIROLAM IM.Mercer Kernel Based Clustering in Feature Space[J]. IEEE Transaction on Neural Networks,2002,13(3):780-784.
[43]张莉,周伟达,焦李成.核聚类算法[J].计算机学报,2002,25(6):587-590.
[44]中国人民大学舆论研究所传播心理实验工作室.读者阅读中文报纸版面的视觉轨迹及其规律—一项基于眼动仪的实验研究[J].国际新闻界,2007,(08):5-19.
[45]田文英.机器学习与数据挖掘[J].石家庄职业技术学院学报.2004,16(6):30-32.
[46]黄良力.多分类支撑向量机综述[J].中国水运(学术版).2006,06(05):72-74.
[47]陈增照,杨扬,何秀玲等.基于核聚类的SVM多类分类方法[J].计算机应用,2007,27(1):47-49.
[2]阎国利,白学军.阅读过程中眼动特征的发展研究[J],心理学探新,2001,21(4):25-28.
[3]沃建中,李琪,田宏杰.不同推理水平儿童在图形推理任务中的眼动研究[J].心理发展与教育,2006,(3):6-10.
[4]阎国利.眼动分析法在广告心理学研究中的应用[J].心理学动态,1999,7(4):50-53.
[5]廖卫华,严由伟,徐子清.视线跟踪技术在人机交互中的研究[J].长春师范学院学报(自然科学版),2005,24(2):130-132.
[6]刘伟,袁修干,庄达民等.眼动测量系统及其在工效学中的应用[J].航空学报,2001,22(5):385-389.
[7]阎国利,白学军.眼动记录法在国外运动心理学研究中的应用[J].心理学动态,1997,5(2):44-48.
[8]韩玉昌.观察不同形状和颜色时眼运动的顺序性[J].心理科学,1997,20(1):40-43.
[9]阎国利,田宏杰,张仙峰.汽车驾驶行为的眼动研究[J].心理科学,2005,28(5):1211-1212.
[10]陈凯,朱钰.机器学习及其相关算法综述[J].统计与信息论坛,2007,22(5):105-112.
[11]刘琴.机器学习[J].武钢职工大学学报.2001,(6):41-44.
[12] Breiman L.Random forests[J].Machine Learning,2001,45(1):5-32.
[13]漆书青,戴海琦,丁树良.现代教育与心理测量学原理[M].南昌:江西教育出版社,1998.
[14] Cortes,C.,Vapnik,V.M.Support Vector Networks[J].Machine Learning.1995,20:273-297.
[15] Dietterich,T.G.An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees:Bagging,Boosting,and Randomization [J].Machine Learning,2000,40:139-157.
[16] Breiman,L.,Bagging Predictors[J].Machine Learning,1996,24(2):123-140.
[17] Agrawal R,Imielinski T,Swami A.Mining association rules between sets of items in large database[A][G].Proc of ACM SIGMOD Conf on Management of Data[C]. Washington,1993:207-216.
[18]刘星沙,谭利球,等.关联规则挖掘算法及其应用[J].计算机工程与科学,2007,29(1):83-86.
[19] P.Dempster,N.M.Laird,A and D.B.Rubin.Maximum Likelihood From Incomplete Data Via the EM Algorithm[J].RoyalStat.Soc.1997,39(1):1-38.
[20] Rosenblatt F . The perceptron : A perceiving and recognizing automaton(Technical report 85-4601-1)[R]. Ithaca NY:Cornell A eronautical Laboratory,1957.
[21] Vapnik V N.Estination of Dependencies Based on Empirical Data.Berlin:Spring-Verlag,1982.
[22] Vapnik V N.The Nature of Statistical Learning Theory,NY:Spring-Verlag,1995.
[23] Cherkassky V,Mulier F.Learning from Data:Concept,Theory and Methods.NY:John Viley&Sons,1997.
[24] Vapink V,Levin E,Le Cun Y.Measuring the VC-dimension of a learning machine.Neural Computation,1994,6:851-876.
[25] Burges C J C.A tutorial on support vector machines for pattern recognition.Data Mining and Knowledge Discovery,1998 2(2).
[26] V Vapnik.The Nature of Statistical Learning Theory[M].New York::Spring Verlag,1995.
[27] Buswell. Growth curves obtained from an objective study of reading habits[J].Journal of Educational Psychology,Volume 13,Issue 7,October 1922, Pages 446-447.
[28] David A. Balota,Alexander Pollatsek, Keith Rayner.The interaction of contextual constraints and parafoveal visual information in reading[J].Cognitive Psychology,Volume 17,Issue 3,July 1985,Pages 364-390.
[29]石金富,曹晓华,王钢,瞿彬.网页布局对视觉搜索影响的眼动研究[J].人类工效学,2008,14(4):1-3.
[30] Chapman P R, Undrwood G. Visual Search of Dynamic Scenes:Event Types and the Role of Experience in Viewing Driving Situations[J].In Underwood ( Eds).Eye Guidance in Reading and Scene Perception,England:Elsevier,1998:369– 394.
[31]张尧庭.广义相关系数及其应用[J].应用数学学报,1978,1(4):312-320.
[32]刘冰.多类SVM分类算法的研究和改进[J].电脑知识与技术,2007,(06):1590-1593.
[33]李昆仑.一种基于有向无环图的多类SVM分类器[J].模式识别与人工智能,2003,16(2):164-168.
[34]曾凡仔,裘正定.一种基于可行域解析中心的多类分类算法[J].复旦学报(自然科学版),2004,43(5):773-776.
[35]唐发明,王仲东,陈绵云.一种新的二叉树多类支持向量机算法[J].计算机工程与应用,2005,(07):24-26.
[36]唐发明,王仲东,陈绵云.支持向量机多类分类算法研究[J].控制与决策,2005,20(07):746-749.
[37]刘至刚等.支撑向量机在多类分类问题中的推广[J].计算机工程与应用.2004,(7):10-13.
[38] CRAMMER K,SINGER Y.On the learn ability and design of output codes for multi-class problems [C].Proceedings of the Thirteenth Annual Conference on Computational Learning Theory,San Francisco,Morgan Kaufman,2000:35-46.
[39] HSU C-W,Lin C-J.A comparison of methods for multi-class support vector machines[J ].IEEE Transaction on Neural Network ,2002,13 (2) :415– 425.
[40]边肇祺,张学工.模式识别[M].第2版.北京:清华大学出版社.2000.
[41] SCHLOPFB,MIKA S, BURGES C,et al. Input space versus feature space in kernel-based methods[J]. IEEE Transaction Neural Networks,1999,10(5):1000-1017.
[42] GIROLAM IM.Mercer Kernel Based Clustering in Feature Space[J]. IEEE Transaction on Neural Networks,2002,13(3):780-784.
[43]张莉,周伟达,焦李成.核聚类算法[J].计算机学报,2002,25(6):587-590.
[44]中国人民大学舆论研究所传播心理实验工作室.读者阅读中文报纸版面的视觉轨迹及其规律—一项基于眼动仪的实验研究[J].国际新闻界,2007,(08):5-19.
[45]田文英.机器学习与数据挖掘[J].石家庄职业技术学院学报.2004,16(6):30-32.
[46]黄良力.多分类支撑向量机综述[J].中国水运(学术版).2006,06(05):72-74.
[47]陈增照,杨扬,何秀玲等.基于核聚类的SVM多类分类方法[J].计算机应用,2007,27(1):47-49.