A Joint Summarization and Pre-Trained Model for Review-Based Recommendation
Round 1
Reviewer 1 Report
The authors have paid attention to the reviews on the Internet for improving an efficiency of a recommendation system. In this manuscript, they propose a joint summarization and pre-trained recommendation model for review-based rate prediction called JSPTRec. Its method was compared with existing seven other methods using four datasets, and they show that the new method is effective.
I recommend its publication to MDPI Information because this manuscript satisfies the level with respect to the presentation, research question, an original approach, comparison and the results. Especially, I appreciate their point of view where information on the reviews can be introduced to a recommendation system. However, the current version has a critical flaw in Methods Section, and thus I must require its substantial revision.
I carefully read that section, but I cannot capture your algorithm. First, r_u and r_i occur confusion. These two different elements must b given different notation. If not, we do not know which definition when we say r_1, for example. As well, C_u and C_i, S_u and S_i, e_i and e_u, and some other pairs are not suitable.
Second, I did not know how the values of S_i and S_u are calculated. This seems to be a crucial point in the review summarization layer and your BERT representation layer. Finally, the function relu is not defined yet in calculating an attention score.
Author Response
Major comments:
- r_u and r_i occur confusion. These two different elements must b given different notation. If not, we do not know which definition when we say r_1, for example. As well, C_u and C_i, S_u and S_i, e_i and e_u, and some other pairs are not suitable.
[authors] We thank the reviewer for this comment. We modified the notations in our method and showed the notations in Table 1. We add superscripts to the symbols to distinguish the symbols on the user side and the item side.
- I did not know how the values of S_i and S_u are calculated. This seems to be a crucial point in the review summarization layer and your BERT representation layer.
[authors] Thanks for the comment. For each review, we calculate the importance of each sentence in it through the TextRank method, and then select the most important K sentences as summary. K can be obtained by µ*|r|, where µ is the proportion of the review summary and |r| is the number of sentences in the review. Similarly, for each item review, the summary is calculated in the same way.
We have a detailed description of how to calculate and in the lines of 113-117 in Section 2.1.
- The function relu is not defined yet in calculating an attention score.
[authors] Thanks for the comment. We have added the explanation and definition of ReLU function in Section 2.3.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper proposes a joint summarization and pre-trained recommendation model for review-based rate prediction. The paper is well-written, the background and the proposed method is clearly presents. However, i still have some minor comments that can help you improve your manuscript.
-please state which implementation you have used for the compared baselines (PMF,NMF ...), if you use open source ones, please provide the code link (e.g. git repo).
-Please add a related work section, where you briefly describe what previous researches have done in this topic. I suggest you add the following papers to your related work section:
(review summarization)
[1] Mabrouk, A., Redondo, R. P. D., & Kayed, M. (2021). SEOpinion: Summarization and Exploration of Opinion from E-Commerce Websites. Sensors, 21(2), 636. MDPI.
[2] Wu, P.; Li, X.; Shen, S.; He, D. Social media opinion summarization using emotion cognition and convolutional neural networks. Int. J. Inf. Manag. 2020, 51, 101978.
[3] Al-Natour, S., & Turetken, O. (2020). A comparative assessment of sentiment analysis and star ratings for consumer reviews. International Journal of Information Management, 54, 102132.
(recommendation systems)
[1] Dhelim et al. Personality-Aware Product Recommendation System Based on User Interests Mining and Metapath Discovery. IEEE Transactions on Computational Social Systems 2020.
[2] Khelloufi et al. A Social Relationships Based Service Recommendation System For SIoT Devices. IEEE Internet of Things Journal. 2020.
[3] Ning et al. PersoNet: Friend recommendation system based on big-five personality traits and hybrid filtering. IEEE Transactions on Computational Social Systems. 2019
Author Response
Major comments:
- please state which implementation you have used for the compared baselines (PMF,NMF ...), if you use open source ones, please provide the code link (e.g. git repo).
[authors] Thanks for the comment. To perform the experiments, we use an open source code on github for PMF and NMF. For other baselines, we use the codes provided by the authors respectively. We have added the code links in the revised version.
PMF, NMF https://github.com/JieniChen/Recommender-System
HFT http://cseweb.ucsd.edu/~jmcauley/code/code_RecSys13.tar.gz
DeepCoNN https://github.com/richdewey/DeepCoNN
MPCN https://github.com/vanzytay/kdd2018_mpcn
D-Attn https://hub.fastgit.org/seongjunyun/CNN-with-Dual-Local-and- Global-Attention
NARRE https://github.com/chenchongthu/narre
- Please add a related work section, where you briefly describe what previous researches have done in this topic. I suggest you add the following papers to your related work section:
(review summarization)
[1] Mabrouk, A., Redondo, R. P. D., & Kayed, M. (2021). SEOpinion: Summarization and Exploration of Opinion from E-Commerce Websites. Sensors, 21(2), 636. MDPI.
[2] Wu, P.; Li, X.; Shen, S.; He, D. Social media opinion summarization using emotion cognition and convolutional neural networks. Int. J. Inf. Manag. 2020, 51, 101978.
[3] Al-Natour, S., & Turetken, O. (2020). A comparative assessment of sentiment analysis and star ratings for consumer reviews. International Journal of Information Management, 54, 102132.
(recommendation systems)
[1] Dhelim et al. Personality-Aware Product Recommendation System Based on User Interests Mining and Metapath Discovery. IEEE Transactions on Computational Social Systems 2020.
[2] Khelloufi et al. A Social Relationships Based Service Recommendation System For SIoT Devices. IEEE Internet of Things Journal. 2020.
[3] Ning et al. PersoNet: Friend recommendation system based on big-five personality traits and hybrid filtering. IEEE Transactions on Computational Social Systems. 2019
[authors] Thanks for the comment. Following the reviewer's suggestion, we added a subsection (Section 4.2) in Related Work, which focused on Review Summarization. We added all the references listed by the reviewer. We also added the following additional references as related work:
[1] Xu, H.; Liu, H.; Zhang, W.; Jiao, P.; Wang, W. Rating-boosted abstractive review summarization with neural personalized generation. Knowledge-Based Systems 2021, 218, 106858. doi:https://doi.org/10.1016/j.knosys.2021.106858.
[2] Su, F.; Wang, X.; Zhang, Z. Review Summarization Generation Based on Attention Mechanism. journal of beijing university of posts and telecommunications 2018.
[3] Nyaung, D.E.; Thein, T. Feature-Based Summarizing and Ranking from Customer Reviews 2015.
[4] Shimada, K.; Tadano, R.; Endo, T. Multi-aspects review summarization with objective information. Procedia - Social and Behavioral Sciences 2011, 27, 140–149.
Author Response File: Author Response.pdf
Reviewer 3 Report
This article describes a research project that is well designed, created, and implemented.
All the required sections are presented with comprehensive details.
The Introduction, Methodology, Results, and Discussions sections are clearly presented and directly related to the objective of this research project.
Recommendations:
I suggest to authors add the following paper to the reference list:
Shakhovska N., Fedushko S., Greguš ml. M., Shvorob I., Syerova Yu. Development of Mobile System for Medical Recommendations. Procedia Computer Science. Volume 155, 2019, pp. 43-50. https://doi.org/10.1016/j.procs.2019.08.010
Author Response
Major comments:
- I suggest to authors add the following paper to the reference list:
Shakhovska N., Fedushko S., Greguš ml. M., Shvorob I., Syerova Yu. Development of Mobile System for Medical Recommendations. Procedia Computer Science. Volume 155, 2019, pp. 43-50. https://doi.org/10.1016/j.procs.2019.08.010
[authors] Thanks for the comment. At the beginning of the introduction, we have added a reference to this paper in the revised version.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I've checked the revised version and I recommend its acceptance. Congrats.
