eye_movements

Dataset used in the tabular data benchmark https://github.com/LeoGrin/tabular-benchmark, transformed in the same way. This dataset belongs to the "regression on numerical features" benchmark. Original description: **Author**: **Source**: Unknown - Date unknown **Please cite**: Jarkko Salojarvi, Kai Puolamaki, Jaana Simola, Lauri Kovanen, Ilpo Kojo, Samuel Kaski. Inferring Relevance from Eye Movements: Feature Extraction. Helsinki University of Technology, Publications in Computer and Information Science, Report A82. 3 March 2005. Data set at http://www.cis.hut.fi/eyechallenge2005/ Competition 1 (preprocessed data) A straight-forward classification task. We provide pre-computed feature vectors for each word in the eye movement trajectory, with class labels. The dataset consist of several assignments. Each assignment consists of a question followed by ten sentences (titles of news articles). One of the sentences is the correct answer to the question (C) and five of the sentences are irrelevant to the question (I). Four of the sentences are relevant to the question (R), but they do not answer it. * Features are in columns, feature vectors in rows. * Each assignment is a time sequence of 22-dimensional feature vectors. * The first column is the line number, second the assignment number and the next 22 columns (3 to 24) are the different features. Columns 25 to 27 contain extra information about the example. The training data set contains the classification label in the 28th column: "0" for irrelevant, "1" for relevant and "2" for the correct answer. * Each example (row) represents a single word. You are asked to return the classification of each read sentence. * The 22 features provided are commonly used in psychological studies on eye movement. All of them are not necessarily relevant in this context. The objective of the Challenge is to predict the classification labels (I, R, C). Please see the technical report for information of eye movements, experimental setup, baseline methods and references: Jarkko Salojarvi, Kai Puolamaki, Jaana Simola, Lauri Kovanen, Ilpo Kojo, Samuel Kaski. Inferring Relevance from Eye Movements: Feature Extraction. Helsinki University of Technology, Publications in Computer and Information Science, Report A82. 3 March 2005. [PDF] Modified by TunedIT (converted to ARFF format) FEATURES The values in columns marked with an asterisk (*) are same for all occurances of the word. COL NAME DESCRIPTION 1 #line Line number 2 #assg Assignment Number 3 fixcount Number of fixations to the word 4* firstPassCnt Number of fixations to the word when it is first encountered 5* P1stFixation '1' if fixation occured when the sentence the word was in was encountered the first time 6* P2stFixation '1' if fixation occured when the sentence the word was in was encountered the second time 7* prevFixDur Duration of previous fixation 8* firstfixDur Duration of the first fixation when the word is first encountered 9* firstPassFixDur Sum of durations of fixations when the word is first encountered 10* nextFixDur Duration of the next fixation when gaze initially moves from the word 11 firstSaccLen Length of the first saccade 12 lastSaccLen Distance between fixation on the word and the next fixation 13 prevFixPos Distance between the first fixation preceding the word and the beginning ot the word 14 landingPos Distance between the first fixation on the word and the beginning of the word 15 leavingPos Distance between the last fixation on the word and the beginning of the word 16 totalFixDur Sum of all durations of fixations to the word 17 meanFixDur Mean duration of the fixations to the word 18* nRegressFrom Number of regressions leaving from the word 19* regressLen Sum of durations of regressions initiating from this word 20* nextWordRegress '1' if a regression initiated from the following word 21* regressDur Sum of durations of the fixations on the word during regression 22 pupilDiamMax Maximum pupil diameter 23 pupilDiamLag Maximum pupil diameter 0.5 - 1.5 seconds after the beginning of fixation 24 timePrtctg First fixation duration divided by the total number of fixations 25 nWordsInTitle Number of word in the sentence (title) this word is in 26 titleNo Title number 27 wordNo Word number (ordinal) in this title 28 label Classification for training data ('0'=irrelevant, '1'=relevant, '2'=correct)本数据集用于表格数据基准测试（tabular data benchmark），与https://github.com/LeoGrin/tabular-benchmark 中采用的预处理方式一致，属于“数值特征回归（regression on numerical features）”基准任务。原始数据集描述如下： **作者**： **来源**：未知 - 日期不详 **请引用以下文献**： Jarkko Salojarvi、Kai Puolamaki、Jaana Simola、Lauri Kovanen、Ilpo Kojo、Samuel Kaski. 从眼动轨迹推断相关性：特征提取（Inferring Relevance from Eye Movements: Feature Extraction）. 赫尔辛基理工大学计算机与信息科学出版物，报告A82，2005年3月3日。数据集可在http://www.cis.hut.fi/eyechallenge2005/ 获取。 竞赛1（预处理数据） 本任务为简易分类任务。我们为眼动轨迹中的每个单词提供了预计算的特征向量，并附带类别标签。 本数据集包含多个任务单元（assignment）。每个任务单元由一个问题及十句话（新闻文章标题）组成。其中一句话为该问题的正确答案（类别C），五句话与问题无关（类别I），另有四句话与问题相关但无法回答该问题（类别R）。 * 特征以列存储，每行对应一个特征向量。 * 每个任务单元是由22维特征向量组成的时间序列。 * 第1列为行号，第2列为任务单元编号，第3至24列共22列为各类特征。第25至27列包含该样本的额外信息。训练数据集的第28列为分类标签：“0”代表无关，“1”代表相关，“2”代表正确答案。 * 每个样本（即每行数据）对应单个单词。任务要求为预测每个读取句子的分类标签。 * 提供的22个特征是眼动心理学研究中常用的特征，在此任务中未必全部适用。 本次挑战赛的目标为预测样本的分类标签（对应类别I、R、C）。 如需了解眼动相关信息、实验设置、基准方法及参考文献，请参阅以下技术报告： Jarkko Salojarvi、Kai Puolamaki、Jaana Simola、Lauri Kovanen、Ilpo Kojo、Samuel Kaski. 从眼动轨迹推断相关性：特征提取（Inferring Relevance from Eye Movements: Feature Extraction）. 赫尔辛基理工大学计算机与信息科学出版物，报告A82，2005年3月3日。[PDF] 本数据集经TunedIT修改（转换为ARFF格式）。 ## 特征说明 带星号（*）标记的列中，同一单词的所有出现对应的特征值完全一致。 | 列号 | 特征名称 | 特征描述 | |------|----------------|--------------------------------------------------------------------------| | 1 | #line | 行号 | | 2 | #assg | 任务单元编号 | | 3 | fixcount | 该单词的注视点（fixation）总数 | | 4* | firstPassCnt | 首次遇到该单词时的注视点总数 | | 5* | P1stFixation | 若该单词所在句子首次被读取时即发生注视，则取值为'1' | | 6* | P2stFixation | 若该单词所在句子第二次被读取时即发生注视，则取值为'1' | | 7* | prevFixDur | 前一次注视的持续时长 | | 8* | firstfixDur | 首次遇到该单词时的首次注视时长 | | 9* | firstPassFixDur| 首次遇到该单词时所有注视点的总持续时长 | | 10* | nextFixDur | 视线首次从该单词移开后的下一次注视的持续时长 | | 11 | firstSaccLen | 首次眼跳（saccade）的距离 | | 12 | lastSaccLen | 该单词上的注视点与下一次注视点之间的距离 | | 13 | prevFixPos | 该单词前一次注视点与该单词起始位置之间的距离 | | 14 | landingPos | 该单词的首次注视点与该单词起始位置之间的距离 | | 15 | leavingPos | 该单词的末次注视点与该单词起始位置之间的距离 | | 16 | totalFixDur | 该单词所有注视点的总持续时长 | | 17 | meanFixDur | 该单词所有注视点的平均持续时长 | | 18* | nRegressFrom | 从该单词发起的回视（regression）次数 | | 19* | regressLen | 从该单词发起的所有回视的总持续时长 | | 20* | nextWordRegress| 若从当前单词的后续单词发起回视，则取值为'1' | | 21* | regressDur | 回视过程中对该单词的所有注视点的总持续时长 | | 22 | pupilDiamMax | 最大瞳孔直径 | | 23 | pupilDiamLag | 注视开始后0.5至1.5秒时的最大瞳孔直径 | | 24 | timePrtctg | 首次注视时长除以总注视点数 | | 25 | nWordsInTitle | 该单词所在句子（标题）包含的单词总数 | | 26 | titleNo | 标题编号 | | 27 | wordNo | 该标题内的单词序号（按顺序排列） | | 28 | label | 训练集分类标签：'0'=无关，'1'=相关，'2'=正确答案 |