Automatic Detection Of Deep Reading And Shallow Reading Behavior Within-document By Using Eye-movements Data Analysis

Reading is an important cognitive process for people to collect information and acquire knowledge.When facing a large amount of information within-document,readers present two different reading behaviors according to their needs and reading engagement,which are deep reading and shallow reading.The main function of shallow reading is to filter the irrelevant text information and find the useful passages,which is the entrance to the deep reading.Deep reading makes a careful reading of the text that is filtered by shallow reading,which means the process is slower and more time-consuming,and the characteristic of line by line reading is presented.At present,the researches about deep reading and shallow reading mainly focus on the concept analysis,feature description and dialectical relationship between deep reading and shallow reading.Effectively automatic detection of deep reading and shallow reading behaviors is of great significance to studying readers' cognitive laws and optimizing their reading experience,but there is no research about the pattern recognition of deep reading and shallow reading,and quantitative researches about the deep reading and shallow reading is quite few.In this case,the paper demonstrates that the eye movement data can reflect the reading behavior within-document,and expounds differences of the deep reading and shallow reading in view of eye movement data.According to the literature research,the paper chooses the eye movements indicators which reflect the deep reading and shallow reading difference,and takes advantage of the difference to distinguish different reading behaviors.By analyzing the relevant literatures,there are some other factors that may affect the chosen eye movement indicators,including the reader factors,document factors and task factors.The paper argues that these factors influence the reading behavior in the reading process,and then affect the eye movement indicators.By controlling the document factors,the paper aims to identify the deep reading and shallow reading behavior when different readers complete different types of tasks.Through the analysis of the current researches about using eye movement data for automatic classification and pattern recognition,the paper demonstrates the feasibility of using eye movement data analysis to automatically detect deep reading and shallow reading behavior,and then puts forward a clustering model based on specific features of eye movement for automatic recognition of deep reading and shallow reading behavior.Finally this paper makes an empirical study on the proposed model.24 different subjects are recruited from Wuhan University to complete three different types of reading tasks respectively,and Tobii eye tracker is used to collect eye movement data at the same time.By input the selected eye movement features,the model outputs identification results.Three experts are invited to judge the identification results and the paper adopts the principles of at least two people of the three.The overall accuracy of section identification is 84.95%and the overall accuracy of the fixation classification is 94.32%,thus the empirical results confirm the validity of this automatic detection method.The contribution of this paper is divided into two aspects:on the one hand,from the perspective of eye movement indicators,this paper provides a new kind of quantitative detection method of deep reading and shallow reading,and it can be automatically unsupervised,which can effectively improve the current problem of distinction between deep reading and shallow reading which is subjective and time-consuming.On the other hand,automatically identifying the deep reading and shallow reading can further reveal the proportion and transformation rule of deep reading and shallow reading,which is of great importance to studying different readers' cognitive mechanism,providing personalized service and optimizing the readers' experience more accurately according to the different reading state,which thus improves the efficiency of information service.