2020 Fiscal Year Annual Research Report
Automatic detection of level of students' engagement
| Project/Area Number |
18K18168
|
|---|
| Research Institution | Okayama University |
|---|
Principal Investigator |
Yucel Zeynep 岡山大学, 自然科学研究科, 准教授 (20586250)
|
|---|
| Project Period (FY) |
2018-04-01 – 2021-03-31
|
| Keywords | Engagement / Attention / E-learning / Behavior |
|---|
| Outline of Annual Research Achievements |
We evaluated the conditional independence relation of the four behavioral variables using entropy distance. We have observed that the duration of blinks t_b and frequency of blinks f_b have the highest rate of independence (0.95), followed by duration of blinks t_b and aspect ratio of the eyes r_o (0.93), and finally duration of blinks t_b and the depth of the user d_io (i.e. distance between the user and the screen, 0.92). Since all values of entropy distance are above 0.90, we assumed that there is sufficient evidence for independence. Subsequently, we built 4 models for each individual behavioral variable and blended them into a single probabilistic estimation of engagement. We have shown that the estimated probability of being engaged increases monotonically with the annotated level of engagement. In addition, we showed that the standard error on the estimations is quite small, suggesting that should there be a larger amount of data, the standard deviation is likely to decrease significantly.
In addition to this integration approach (i.e., employing the set of all variables), we can also applied a “differential approach,” where we remove one variable at a time from the input set, as another means to evaluate the sensitivity of the method to each individual variable. In this way, we confirmed that d_io provides the largest amount of contribution followed by r_o, f_b and t_b, respectively.
|
