Optimization Algorithms for Detection of Social Interactions

Round 1

Reviewer 1 Report

The first sentence of the abstract has a singular-plural conflict: "problem" -> "problems".

The Oxford comma only applies in lists of three or more elements: "strategy, and related" -> "strategy and related". Similar errors occur further along in the manuscript. Please review and revise the punctuation.

"As assert in [5]" -> "As asserted by Brandes et al. [5]" (citations are not words).

On line 43, Q is missing the mathmode: $Q$.

The word "together" on line 90 should probably be "together with the".

"subsection 2.2" -> "Section 2.2"

"randomly selects a vertex from the solution and reassigns it" -> "randomly select a vertex from the solution and reassign it"

"randomly selects a cluster" -> "randomly select a cluster"

"will be then moved in the cluster" -> "are then moved to the cluster"

"doesn’t exist" -> "does not exist" (good style in scientific writing is not to abbreviate these constructs)

"a new community will be created" -> "a new community is created" (similarly, scientific writing prefers the present for presenting proposed methods)

"This ensure" -> "This ensures"

"the first one" -> "the former"

"the last one" -> "the latter"

"choices, that refine" -> "choices that refine"

The first paragraph on page 5 is too long and could use some restructuring to improve legibility.

"in table 1" -> "in Table 1" (similar style errors repeat later on)

All the networks in the experiments are extremely small.

Right-align the numerical columns (second and third) in Table 1.

The title inserted in the figure itself in Figure 1 is redundant as the caption includes the information already. Please remove it.

"From figure 1" -> "From Figure 1"

You can NEVER use the fitness measure YOU EXPLICITLY OPTIMIZE in your algorithm to compare your results to OTHR algorithms that explicitly optimize OTHER measures and expect the comparison to be meaningful in any sense. It is very important to clarify whether all the seven algorithms actually aim to maximize modularity, defined the same exact way than you implemented it. Otherwise Table 2 will not mean anything at all.

Since you are using networks for some of which ground truths are available, comparing to those would be more informative.

Also, runtimes need to be reported as well for a comparison to be useful. Also the variability of the modularity in all methods that are stochastic has to be reported; mean and standard deviation over 30 runs or so if it is roughly normally distributed, box plots or violin diagrams otherwise.

Section 4 contains no discussion of future work.

 

Author Response

All our responses to the comments of Reviewer #1 (point by point) can be found in the attached pdf file.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this article, the authors propose 2 Immunological Algorithms for the community detection problem. They aim at showing the efficiency and robustness of the proposed algorithms compared with the state of the art.
Overall, the work is interesting and the theme is important. It is well-written but a review would be welcome.

Overall, the work three major flaws.

First, It is not clear for readers what are the benefits of the new approaches. The state of the art already presents high modularity. New approaches demand a high number of interactions to achieve Louvain modularity. What are the gains?
In this sense, the authors must carefully discuss the evaluation metrics. Besides modularity, one could address the time or memory consumption.

Second, what are the approaches limits and boundaries of the new approaches? In other words, they are well applied in which context? For example, some methods outperform Louvain in a high dense network (or in a low dense network).

Finally, the authors evaluate their approach in very small networks. Nowadays, most of the networks present many more edges and vertices and edges. For example, some recent works address the following networks (from low to high dense networks)

Amazon 334,862 - 925,872 (vertices and edges)
Youtube 3,223,642 - 9,296,202
Flickr 2,302,924 - 33,140,018
Live Journal 5,204,174 - 49,174,464



In what follows, I point minor suggestions:

1) Using equations in the introduction section is weird. I think there are more comfortable ways of explaining this topic in the introduction, saving equations for core sections.

2) The state of the art is presented as if the preexistent techniques are all good. Hence, why are your proposals necessary? First, you need to identify a weakness in the state of the art. Then, you show your proposal as an improvement or a solution.

3) I miss the numbers in the results section. For example, Technique A outperforms technique B in x%, which leads to an improvement of ???. Why put an equation in the introduction if you are being so abstract in the results section.

4) Maybe you should replace the table for a graphic representation.

5) You finish the results section with the sentence: "OPT-IA and HYBRID-IA perform very well on networks (large and small) considered for the experiments, and are competitive with the state-of-the-art in terms of efficiency and robustness". Ok... but why do I should use your technique instead of the stable and extensively tested preexistent ones.

6) Line 248 – missing reference

Author Response

All our responses to the comments of Reviewer #2 (point by point) can be found in the attached pdf file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Line 57: In section 2 -> In Section 2

Line 61: in the subsections -> In Sections (this is established style, the numbering implicitly indicates those are subsections)

Line 115: subsection -> Section

Line 253: In figure -> In Figure

Line 273: of the figure 1 -> of Figure 1

In Table 2, please underline the best value on each row.

 

Author Response

We are very grateful to the reviewer for having allowed us to significantly improve the manuscript, and our future research with his useful and challenging comments and suggestions.

Attached the responses to the reviewer's comments.

Author Response File: Author Response.pdf

Reviewer 2 Report

Overall, the authors addressed my questions from the previous review round. I noticed all along with the text the new argumentation, as they provided in the response letter.

 

There are still two points authors must address before the paper acceptance.

 

First, the existence of bigger networks is not “in the literature”, but in real life. It should not be difficult to develop it, once the method is already created. To compare the existent algorithm with several others, it is not the size of the dataset that matters, but a way to reproduce the algorithm (or code availability). One should be able to use any existing dataset to test it. 

 

In this sense, I wonder if the current method is able to deal with larger networks. The authors must evaluate the algorithm scalability. It seems that it does not scale.

 

Second, when previously mentioned, “authors must carefully discuss the evaluation metrics. Besides the modularity, one could address the time or memory consumption” I was indicating: authors must provide the analysis of the complexity of the proposed algorithm (since the focus of the current journal is algorithm).

 

Author Response

We are very grateful to the reviewer for having allowed us to significantly improve the manuscript, and our future research with his useful and challenging comments and suggestions.

Attached the responses to the reviewer's comments.

Author Response File: Author Response.pdf