Embodied Emotion Recognition Based on Life-Logging
Round 1
Reviewer 1 Report
This is an interesting study with potential contribution in the field. I would like to commend the authors to operationalize emotional with a theoretical driven approach using Russell's circumplex model of affect. However, there are some area for improvement:
There is a lack of details of how the data on participants' emotion was collected. What is the questionnaire that was used? What are the items? Is the scale reliable? The information should be reported in the manuscript. In the page 4, the authors wrote: Finally, the hypothesis of this study that significant causalities, as a result of multiple regression, differ depending on the emotions is verified by ANOVA" I doubt the correlational design that was implemented by the authors can determine any causality. Moreover, using multiple regression does not imply cause and effect. The authors should be careful in their interpretation throughout the manuscript. Limitations of the study should be expanded further. It will be good to elaborate further the data pre-processing procedure. It will be good to summarize and describe all the 31 variables in a table.
Author Response
"Please see the attachment."
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The article concerns the methods of recognizing emotions. The authors draw attention to the lack of research that would take into account physiological activity, behavior and environmental variables at the same time. They conducted a study in ecological conditions, in which the subjects reported their emotions, while collecting heart rate measurements and ambient sounds. The idea is undoubtedly interesting and novel, since there is a lack of research into the recognition of emotions that would take into account all three factors at the same time in ecological conditions.
However, I have concerns about the description of the method and the analysis of the results. Firstly, the way in which the data is collected has not been clearly explained. The respondents wore a PPG measuring device and used the phone for 5 hours a day. Were these hours fixed? Was it work time or rest time? Did they wear the device continuously? Was it somehow verified (e.g. by checking the continuity of measurements)? Did it not interfere with everyday activities (e.g. washing hands or job-related tasks)? Was the assessment of emotions made every hour initiated by the respondents, i.e. did they have to start the application and assess the emotions, or whether there were any notifications from application? What if e.g. someone had a long telephone conversation or drove a car and was unable to make a timely assessment?
The biggest problem in my opinion is the analysis of the results. The main result presented in the article are the diagrams of causal relationships between variables. But on the basis of multiple regression and ANOVA model (which is in fact a specific case of regression analysis) it is not possible to imply causation. In the proposed study this is particularly important as it is not an experimental study, but a "correlational" study: you collect several variables at the same time, but you do not manipulate anything. In fact, the results would still be valuable if there will be information only about the correlations between variables instead of the causal relations.
I also have several concerns about the presentation of the results itself. The most important ones:
(a) in lines 235-237 you stated "The level of arousal was classified into three levels based on the average of the assessment. The data larger than the average were classified as arousal, the average was neutral, and the data smaller than the average were classified as relaxation." -- why that way? Why you do not used the middle value on likert scale as a neutral value? What if average was e.g. "3.4"? All values below (up to 3.99) were classified as relaxation and all above (4.01 and more) were classified as arousal, so only 4.00 was neutral? In this way, the number of neutral cases is very small.
(b) Tables 2 and 4: How do I understand the values in this table? For example, the first value: -0.015±0.259 is Mean and SD for the dependent variable for arousal classified as "Arousal"?
(c) Tables 3 and 5: What is in the columns marked as post-hoc? Differences between the averages? If so, for which levels? Where is the information about whether a given difference (a given post-hoc test) was statistically significant or not? What tests were performed as post-hoc tests?
(d) Tables and schemes have been badly formatted. Schemes have too large fonts and overlapping arrows. Tables are placed on the margins.
I suggest you work through the analysis of the data again and focus on showing the relevant correlations.
Author Response
"Please see the attachment."
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have sufficiently addressed my concerns. This is a good paper and I hope to see it published. Congratulation to the authors!
Reviewer 2 Report
The article has been significantly improved and now it is fine. All the problems are properly addressed in the text, I have no further comments to make.
As I stated in previous review, I find it interesting and original because the authors have conducted a study in which they take into account physiological activity, behavior and environmental variables at the same time. This is important to perform ecological research like that, and there is still a lack of such studies.
