TY - GEN
T1 - Predicting multi-document comprehension
T2 - 20th International Conference on Artificial Intelligence in Education, AIED 2019
AU - Nicula, Bogdan
AU - Perret, Cecile A.
AU - Dascalu, Mihai
AU - McNamara, Danielle S.
N1 - Funding Information:
Acknowledgments. This research was supported by the ReadME project “Interactive and Innovative application for evaluating the readability of texts in Romanian Language and for improving users’ writing styles”, contract no. 114/15.09.2017, MySMIS 2014 code 119286, the FP7 2008-212578 LTfLL project, the Institute of Education Sciences (R305A180144 and R305A180261), and the Office of Naval Research (N00014-17-1-2300).
Publisher Copyright:
© Springer Nature Switzerland AG 2019.
PY - 2019
Y1 - 2019
N2 - Theories of discourse comprehension assume that understanding is a process of making connections between new information (e.g., in a text) and prior knowledge, and that the quality of comprehension is a function of the coherence of the mental representation. When readers are exposed to multiple sources of information, they must make connections both within and between the texts. One challenge is how to represent this coherence and in turn how to predict readers’ levels of comprehension. In this study, we represent coherence using Cohesion Network Analysis (CNA) in which we model a global cohesion graph that semantically links reference texts to different student verbal productions. Our aim is to create an automated model of comprehension prediction based on features extracted from the CNA graph. We examine the cohesion links between the four texts read by 146 students and their (a) self-explanations generated on target sentences and (b) responses to open-ended questions. We analyze the degree to which features derived from the cohesive links from the extended CNA graph are predictive of students’ comprehension scores (on a [0 to 12] scale) using either (a)Â students’ self-explanations, (b)Â responses to comprehension questions, or (c)Â both. We compared the use of Linear Regression, Extra Trees Regressor, Support Vector Regression, and Multi-Layer Perceptron. Our best model used Linear Regression, obtaining a 1.29 mean absolute error when predicting comprehension scores using both sources of verbal responses (i.e., self-explanations and question answers).
AB - Theories of discourse comprehension assume that understanding is a process of making connections between new information (e.g., in a text) and prior knowledge, and that the quality of comprehension is a function of the coherence of the mental representation. When readers are exposed to multiple sources of information, they must make connections both within and between the texts. One challenge is how to represent this coherence and in turn how to predict readers’ levels of comprehension. In this study, we represent coherence using Cohesion Network Analysis (CNA) in which we model a global cohesion graph that semantically links reference texts to different student verbal productions. Our aim is to create an automated model of comprehension prediction based on features extracted from the CNA graph. We examine the cohesion links between the four texts read by 146 students and their (a) self-explanations generated on target sentences and (b) responses to open-ended questions. We analyze the degree to which features derived from the cohesive links from the extended CNA graph are predictive of students’ comprehension scores (on a [0 to 12] scale) using either (a)Â students’ self-explanations, (b)Â responses to comprehension questions, or (c)Â both. We compared the use of Linear Regression, Extra Trees Regressor, Support Vector Regression, and Multi-Layer Perceptron. Our best model used Linear Regression, obtaining a 1.29 mean absolute error when predicting comprehension scores using both sources of verbal responses (i.e., self-explanations and question answers).
KW - Cohesion network analysis
KW - Comprehension modeling
KW - Machine learning
KW - Multi-document comprehension and integration
KW - Natural language processing
UR - http://www.scopus.com/inward/record.url?scp=85068328134&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85068328134&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-23204-7_30
DO - 10.1007/978-3-030-23204-7_30
M3 - Conference contribution
AN - SCOPUS:85068328134
SN - 9783030232030
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 358
EP - 369
BT - Artificial Intelligence in Education - 20th International Conference, AIED 2019, Proceedings
A2 - Isotani, Seiji
A2 - Millán, Eva
A2 - Ogan, Amy
A2 - McLaren, Bruce
A2 - Hastings, Peter
A2 - Luckin, Rose
PB - Springer Verlag
Y2 - 25 June 2019 through 29 June 2019
ER -