Customer Loyalty Improves the Effectiveness of Recommender Systems Based on Complex Network
Round 1
Reviewer 1 Report
The paper shows the in-depth knowledge of the authors in the field and their commitment to the application. Topic selection is actual, and the methodological background is excellent and progressive. There are several details explained in the text about the methods and the procedure of the investigation. However, the results and conclusions are pushed in the background. This is why I recommend a major revision of the paper before publishing it.
I cannot decide which is more critical: validation of the methodology and the procedure or solving the practical problem. Depending on the purpose, the empirical dataset must be used just as an example, or the results, discussion, and conclusion session must be emphasized while overshadowing the exhausting presentation of the methodology.
Another note is that there are some abbreviations in the text. E.g., RFML is not described at first occur. I suggest adding a list of abbreviations.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The objective of the article is to consider the impact of customer loyalty on the accuracy of recommendation algorithm that measures the similarity shops to customers based on their historical purchase records. It uses the DeepWalk algorithm to search a platform of a real estate company in Malaysia.
The results show that there is an improvement in the effectiveness of the recommendation when consumers have “high levels” of fidelity (measured through RFML).
It is a complex paper to assimilate. It is necessary to value very positively the proposed method used, combining existing algorithms. However, I think you should explain why you use the experiment values ​​θ = 70, t = 50, α = 0.3 on line 461 when you apply SVDW. Possibly they have done a simulation to set the values ​​that provide the best results, but they must indicate it.
I think the comparison of the application of DeepWalk in clusters (cultormer loyalty levels) other than 4 should also be commented, because for them the results are worse. Cluster 4, in relative terms, has an intermediate RFML level (see table 2). An interpretation of the result of the DeepWalk algorithm could be that the results are better when the consumer has an intermediate level of fidelity. This should be clarified.
This would imply that the most recent data weighs more than the oldest. The context of the consumer may have changed over time, as well as his preferences, which would change the values ​​of F, M most recent in the model. Therefore, I propose to give less importance to L and to limit the fidelity estimate to a recent period; for example, the last year or the last two years at most and compare current results with those obtained using this "newer loyalty" version.
The statement they make on lines 31 and 32, I think is not correct, at least as it is written: “… according to customers’ preferences will not change due to changes in the market or advertising of other products”.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors mad relevant changes in the text. They have followed and answered my critical notes. However, I cannot se the final editing in this corrected form, I assume that is correct.
I can recommend the publication.
Reviewer 2 Report
This line of work is very interesting and I encourage youto continue proposing new algorithms and recommendation
procedures. With the future in mind, I think that the RFML
model can be more predictive if we consider different weights
associated with each component.
