FSCR: A Deep Social Recommendation Model for Misleading Information
Round 1
Reviewer 1 Report
The paper entitled "FSCR: A Deep Social Recommendation Model for Misleading Information" describes a recommendation model (FSCR model) which fuses users' explicit information and users' side information in order to identify misleading ratings and based on the authors' statement the proposed model can effectively improve the accuracy and robustness of recommender system.
At first, based on the reference to the actions of "emergency consumption" that lead to the phenomenon of user preferences deviation sounds and interesting problem to face when dealing with recommender systems' accuracy.
I find the paper well organized and easy to follow, and the description of the proposed approach is adequate while the experimental setup and results are covered in detail. The goal of the task is well defined from the beginning and the sections that follow explain each step of the pipeline of analysis to a good extend. The problem formulation including the model architecture, the embedding layer as well as the matrix factorization, the social influence diffusion, and the prediction layer are fully covered, explaining each part of the FSCR model architecture. To verify the effectiveness of the proposed model, authors examined four well-known datasets evaluated among various baseline algorithms and the recommendation results are promising.
Regarding the description of the proposed model and its evaluation, I have no point to raise. But few issues that need to be addressed are the reflection of the main contribution of this work in the abstract which is currently not clear enough and another proofread should be made to fix major spelling or syntax errors, since there are paragraphs in section 4 that need heavy refactoring. Even parts of the Journal's template (like the Journal title) are missing.
Finally, there are no limitations or restrictions stated for the proposed model so I would like to see the authors' position regarding the possible limitations of their approach and a statement regarding the applicability and generalization of this model.
Author Response
Response to Reviewer 1 Comments
We sincerely thank the editor and all reviewers for their valuable feedback that we have used to improve the quality of our manuscript named “FSCR: A Deep Social Recommendation Model for
Misleading Information”. The reviewer comments are laid out below in italicized font and specific concerns have been numbered. Our response is given in normal font and changes/additions to the manuscript are given in the yellow text.
Comments 1:But few issues that need to be addressed are the reflection of the main contribution of this work in the abstract which is currently not clear enough
Response: We have rewritten the abstract and added the description of our work. In fact, our model is a recommender integrating multi-domain information. In order to deal with misleading information, we add an abnormal user identifier to the recommender. After identifying the abnormal user, we do not directly delete all the records of the abnormal user but update his user representation.
We added to page1, Line 15 - line 19
Comments 2：since there are paragraphs in section 4 that need heavy refactoring.
Response: According to your comments, we have revised the manuscript extensively. If there are any other modifications we could make, we would like very much to modify them and we really appreciate your help. We hope that our manuscript could be considered for publication in your journal. Thank you very much for your help.
We updated to page 9-page 11.
Comments 3：Finally, there are no limitations or restrictions stated for the proposed model so I would like to see the authors' position regarding the possible limitations of their approach and a statement regarding the applicability and generalization of this model.
Response: We accept your comments and we describe the limitations of our model at the end of section 4.
We updated to page 11.
In the end, all the changes are marked in PDF. Thank you again for your positive comments and valuable suggestions to improve the quality of our manuscript. If you have any comments, please don't hesitate to contact me.
Best regards,
Depeng Zhang
1550736190@qq.com
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors investigate explicit and user side information in a deep learning model to deal with the problem of misleading information.
The proposed approach is interesting but there are some points that the authors have to better discuss.
The authors should be better described the novelties of their approach with respect to existing ones. In particular, the author should discuss limitation and cons that their approach aims to overcome at the end of the Related Works section. Furthermore, the authors should provide more details and discussion about the obtained results. The Discussion section also needs to be improved by analyzing the outcome of evaluation section.
I suggest to analyze also more recent approaches about the examined topics. In particular, I suggest the following papers to investigate multimedia content in the diffusion and recommendation process:
1) An emotional recommender system for music. IEEE Intelligent Systems
2) Multimedia story creation on social networks. Future Generation Computer Systems, 86, 412-420.
Finally, I suggest to perform a linguistic revision.
Author Response
Response to Reviewer 2 Comments
We sincerely thank the editor and all reviewers for their valuable feedback that we have used to improve the quality of our manuscript named “FSCR: A Deep Social Recommendation Model for
Misleading Information”. The reviewer comments are laid out below in italicized font and specific concerns have been numbered. Our response is given in normal font and changes/additions to the manuscript are given in the yellow text.
Comments 1:The authors should be better described the novelties of their approach with respect to existing ones. In particular, the author should discuss the limitations and cons that their approach aims to overcome at the end of the Related Works section.
Response: We can not clearly describe the changes of our model compared with the existing model in the first draft. There are two ways to deal with the misleading information in the traditional recommendation system. One is to identify and remove the misleading information before training, which will waste a lot of time and computing resources. The other is to improve the robustness of the recommender system by introducing confrontation in the training set. This method will reduce the performance of the recommender system due to the introduction of false scoring. Our model is to integrate a variety of user information (user rating information, user social attribute information, user trust information) to carry out user modeling. At the same time, in order to deal with the misleading information in the recommender system, we also introduce an abnormal user identifier in the model. Different from the traditional behavior of clearing all the ratings of abnormal users directly, we update the user representation of abnormal users. This method can ensure the integrity of the data set. At the same time, our model is superior to the existing recommendation model in score prediction performance because there is no introduction of confrontation.
We added to page 4, Line 146 - line 160
Comments 2：Furthermore, the authors should provide more details and discussion about the obtained results. The Discussion section also needs to be improved by analyzing the outcome of the evaluation section.
Response: Our experiment is divided into four parts: the performance of the FSR model, the performance of the FCR model, the performance of the FSCR model, and the robustness of the FSCR model. And we reorganized and analyzed our experimental results. From the experimental results, the performance of our model is better than that of the traditional matrix factorization model. At the same time, in order to prove the robustness of our model, we also carried out corresponding experiments. Due to the time and equipment, we were not able to clearly describe our experimental results in the table. However, in the experimental part, the fourth part, we have described our experimental results in detail. If you think there are still problems, please do not hesitate to tell me.
We updated to page 9-page 11, Line 267 - line 305
Comments 3: I suggest to analyze also more recent approaches about the examined topics. In particular, I suggest the following papers investigate multimedia content in the diffusion and recommendation process:
1) An emotional recommender system for music. IEEE Intelligent Systems
2) Multimedia story creation on social networks. Future Generation Computer Systems, 86, 412-420.
Response: We sincerely appreciate the valuable comments. We have checked the literature carefully and added more references to the INTRODUCTION part in the revised manuscript.
We added to page 1, Line 22 - line 28
Comments 4: I suggest to perform a linguistic revision
Response: According to your comments, we have revised the manuscript extensively. If there are any other modifications we could make, we would like very much to modify them and we really appreciate your help. We hope that our manuscript could be considered for publication in your journal. Thank you very much for your help.
In the end, all the changes are marked in PDF. Thank you again for your positive comments and valuable suggestions to improve the quality of our manuscript. If you have any comments, please don't hesitate to contact me.
Best regards,
Depeng Zhang
1550736190@qq.com
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I think that the authors have addressed all my concerns.
