Cognitive Similarity-Based Collaborative Filtering Recommendation System

Round 1

Reviewer 1 Report

This manuscript has proposed a new algorithm on collaborative filtering recommendation systems using cognitive similarity of the users based on five features (genre, title, actors, director, and plot) having movie products. The reviewer positively assesses this manuscript because it is well-structured presentation, is simply understandable, proposes a new algorithm satisfying originality, and has sufficient discussion. Despite of this assessment, the manuscript has three major points which should be substantially improved for the publication.

1) Methodological originality.
This manuscript considers the context of items. However, this point of view is not so new. For example, Villegas et al. (2018) [1] has proposed a context-aware recommender system in the same direction. Also, this manuscript uses the genre information of items. However, many papers have adopted the genre information so far. For example, Fremal and Lecron (2017)[2] have used item metadata information (item genres). Barman et al.(2019)[3] have calculated the similarity among the items based on the genre of items. Hasan and Roy (2019)[4] have incorporated items' genre data. The authors should discuss their methodological originality compared with many related papers including the above references.

2) Evaluation methods
Normally, some metrics are used to evaluate a recommendation system. As well as many related papers, this manuscript also uses Precision, Recall, the F1 value, and the MAE value. Why did the author not use the RMSE value? The author should either calculate the RMSE value or discuss the reason it was not used.

3) Extension
The main point of the algorithm this manuscript has proposed was using five features of movie products (genre, title, director, plot, actors). Can this approach be extended to other fields? For example, a book has no actors, a toy has neither plot and actors. The title of this manuscript is not restricted to movie product. The revised manuscript should be required sufficient discussion about that.

[1] https://www.sciencedirect.com/science/article/abs/pii/S0950705117305075
[2] https://doi.org/10.1016/j.eswa.2017.01.031
[3] https://dl.acm.org/doi/abs/10.1145/3316615.3316732
[4] https://www.mdpi.com/2504-2289/3/3/39

Author Response

Our replies has been uploaded by attaching the file. 

Author Response File: Author Response.docx

Reviewer 2 Report

The introduction is informative. However, the part that provides the contribution overview (lines 70-100) needs some simplification/improvement to be more understandable.

I wonder how the readers are expected to understand “Table 1” in the introduction. It’s great the idea to provide an overview of the result. But it should be properly and exhaustively explained in concept first of all. In other words, is the meaning of table 1 clear for readers at that stage?

The Related Work section should be somehow structured and should provide a more consistent literature review.

Section 3 is very hard to read and understand. At the beginning, the notation results not very clear and it affects probably the whole understanding. I really hope authors may introduce some simplification to facilitate the understanding of the method. Personally, I got completely lost reading section 3.

As you are submitting the paper to a multi-disciplinary journal, it would be good to make it understandable for a wide audience. In general, it should be as self-contained as possible. Please try to provide a simple definition for the different concepts (e.g. “cosine similarity”).

I’m not sure I have correctly understood the meaning, the value and the purpose of section 4.

The platform for assessment seems to be not accessible without login. I wonder whether you may grant the access to reviewers. I cannot create an account because it requires an email address which is evidently an identifiable data. The description of such a platform includes a number of technical details on the technology adopted which is pretty standard for Web platforms and doesn’t actually add value. On the contrary, some screenshot or anyway visualization could better overview the high-level features in this case.

The claim in the paper is serious:

"Compared with the Pearson Correlation, our method more accurate and achieves improvement over the baseline 11.1% in the best case. The result shows that our method achieved consistent improvement of 1.8% to 3.2% for various neighborhood sizes in MAE calculation. This indicates that our method improves recommendation performance."

I wonder whether you may provide as more information as possible on the experiments conducted in order to support it.

Author Response

Our replies have been uploaded by the file attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I think that the revised version has been substantially improved for the publication.

Author Response

Thank you for your review.

Reviewer 2 Report

According to the text highlighted, I don't see any change in the introduction and just a minor adding in the related work section. Not clear to me the answer to the comment 2 as well as to the comment 4, as I cannot appreciate any concrete change/adding in the paper. In my humble opinion, not too much effort in the additional explanations provided either. 

Author Response

Thank you very much for your time to review. The revision letter was found in the file attachment.

Author Response File: Author Response.docx