An Improved Link Prediction Approach for Directed Complex Networks Using Stochastic Block Modeling
Round 1
Reviewer 1 Report
The paper lacks clarity and details in discussing the proposed three-process method. Suggest describing the method in a more comprehensive fashion rather than discussing each process separately. A figure may be justified to show the three processes and their connections.
Lines 95-98. The statement:” Hierarchical clustering is a multistage process. Firstly the network is divided into clusters in each iteration, where the most dissimilar nodes are separated. Secondly, the similar nodes are agglomerated together until the termination criteria.” This statement is very confusing, especially the paper continues to describe the bottom-up hierarchical clustering method.
Th list of abbreviations seems redundant since the they are defined in the later contexts they appear.
There are many notational errors (e.g., lines 169, 170, 192, 194, 207, and Eq. 5). Some obvious typos in superscripts or subscripts.
Line 240, what do you mean by “Some model parameters are selected while calculating correlation to maximize the structure’s likelihood.”? This requires further clarification.
It appears the link prediction is based on a probabilistic model. However, there is no details on how the model was estimated at the first place. Eq. 5 deserves additional elaboration.
Author Response
Thank you sir for spending time in reviewing our paper and pointing out the mistakes. We have addresses all the mistakes and queries pointed out to us. The majors changes that we addressed are
(i)We have elaborated the three step process with elaborate explanation examples and necessary figures.
(ii) The abbreviations are cross checked and corrected.
(iii)The notations in the equations are corrected provided with supporting definitions where-ever required.
(iv)The equation for probabilistic approach is corrected and re written.
Reviewer 2 Report
The paper studies an interesting problem about link prediction on complex directed graphs. They proposed a pipeline to package several algorithms to generate the results.
However, the paper still has some following issues to be addressed.
- There is no baseline algorithms compared in the experiments. It is hard to judge the performance of the proposed algorithm in terms of the improvement over existing algorithms.
- The presentation needs to be improved. For example, definition 1 is not very rigorous. What is m? What are l_i and l_j? Those notations are not explained.
- Can the authors justify the pipeline first to find the influencers and do clustering on those influencers? Is this because of efficiency consideration or accuracy consideration? Maybe some experiments can also be provided.
Author Response
Our sincere thanks in reviewing our paper and pointing out the queries. We have addressed the queries that are being raised.
(i) The major link prediction algorithms use either local or global structural information, we have compared and found that the proposed method outperforms the existing algorithms.
(ii) We have rewritten and improved our article by providing necessary definitions, figures and explanations.
(iii) We have also written proper justifications to our three phased process.
(iv) The article has been proof read by our university proof reader and changes are introduced in the presentation styles
Reviewer 3 Report
（1）There are a large number of unnecessary uppercase and space in the context
(2)Ln 11 We observe that the proposed approach outperformed the existing state of the art, In the figure7 the improvement is insignificant
(3)Ln 157-158 what's the contribution of the unnamed list?
(4) Ln 162, what's the meaning of a future link? should it be a hidden link?
(5) Figure 2 is unnecessary
(6)Ln 58-59, are you sure you propose the SBM method?
Author Response
Thank you sir for reviewing the article and pointing the errors. The article has been now proof read and unnecessary upper cases and the grammatical mistakes are corrected.
The article is rewritten by including necessary definitions, figures and illustrations to support the claims.
Round 2
Reviewer 2 Report
The authors have address my comments.
