Is the Part Greater than the Whole : A Comparison of Affective and Behavioral Models Derived from Feature Sub-sets

Philippine Information Technology Journal

Rodrigo, Ma. Mercedes T.¹  Rodriguez, Ramon L.¹  Villaflor, Kathrina Blanca V.¹  Castillo, Vincent Raymond C.¹ 

关键词: Biometrics;    Log file analysis;    Affect;    Data Mining;    Behavior;    Intelligent tutoring systems;    Aplusix;   

DOI  :  10.3860/pitj.v3i2.2572

学科分类：计算机科学（综合）

来源: Philippine Society of Information Technology Educators

【 摘 要 】

Affective computing is computing that relates to user emotion, feelings and motivation. One core problem that it tries to address is the automatic detection of user affect. In this paper, attempts were made to develop models of affective and behavioral states that users exhibit while using Aplusix, an intelligent tutoring system for Algebra. We gathered both user interaction log data and biometrics data from first year Information Technology students at the Mapua Institute of Technology. Both logs were synchronized, cut into time frames, and labeled following rules that we formulated for identifying the specific states of interest. We then used logistic regression and decision tree algorithms to model student affect and behavior based on three feature sets – data from the log files, data from the biometrics logs and a combination of the two. We attempted to determine which feature set was able to produce a more accurate model of affective states of boredom, flow and confusion, and on-task and off-task behavior. We found that logistic regression produces a more accurate model using the feature set from the log files. However, there are no changes in its accuracy when compared to the logistic regression model produced using the combined feature set.

【 授权许可】

Unknown