Deep Reinforcement Learning for Network Dismantling: A K-Core Based Approach
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have presented a deep reinforcement learning-based approach for Network Dismantling. Here are my comments:
1. Can the authors provide statistical properties of each dataset?
2. How are the synthetic graphs were generated and what parameters values were used?
3. Why did the author choose Barabási–Albert (BA) model to generate synthetic graphs? Do the real-world graphs used in this paper also follow BA model?
4. How is the test set chosen? How many times each experiment was performed?
5. How are the hyper parameters chosen?
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
please find attach
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe following are the suggestions on the manuscript "Deep Reinforcement Learning for Network Dismantling: A K-2 Core Based Approach".
The lines 105-106 are written twice. "Meanwhile, approaches to network dismantling can be categorized into planning based, metaheuristic, and machine learning methods. approaches to network dismantling can be categorized into planning-based, metaheuristic, and machine learning methods."
In my opinion, using metaheuristic methods like variable population memetic search and tabu search over conventional approaches for network dismantling must be justified with solid literature work? Specifically, addressing that how do they take use of local structural properties and dynamically modify population size? Line number 114 needs to be modified.
What is the exact method by which SmartCore formulates the K-core problem as a Markov Decision Process (MDP) and how it uses graph neural networks and reinforcement learning to create heuristic methods that minimize the accumulated 2-core size during the dismantling process? It's missing in the Introductory part on the manuscript. Could the authors address the potential computational complexity of GraphSAGE and deep Q learning algorithms, particularly in handling large-scale graphs?
It is also suggested that to make sure it is resilient and generalizable, its performance on different datasets has to be assessed.
What would be the performance of SmartCore in dynamic environments when node attributes change over time or the network structure change? However, authors mention the use of a virtual node to enhance the algorithm's representation capabilities, which could potentially help address challenges posed by dynamic environments but what specific graph-level features does the virtual node capture?
The future directions of the proposed work is misinformation.
Comments on the Quality of English LanguageThe English Language seems appropriate.
Author Response
please find attach
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI have no further comments.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe analytical and mathematical exploration in the manuscript can be improved. English writing should be polished regarding grammatical, lexical, and punctuation points.
Comments on the Quality of English LanguageEnglish writing should be polished regarding grammatical, lexical, and punctuation points.