--======== Review Reports ========-- The review report from reviewer #1: *1: Is the paper relevant to ICDM? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [_] 4 (Innovative) [X] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [_] 4 (High) [X] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [_] 4 (Good) [X] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to ICDM users and practitioners? [_] 3 (Yes) [X] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [_] 2 (High) [X] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [_] 4 (Accept: a regular paper) [_] 3 (Weak Accept: a short paper) [X] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Summary of the paper's main contribution and impact This paper aims to detect overlapping network communities based on network structure and node attributes. The proposed method CESNA extends the network structure-based approach BigCLAM. CESNA outperforms comparison partners in accuracy and scalability. *9: Justification of your recommendation The novelty, improvement, and significance of this research are limited. Some discussions are not very precise. *10: Three strong points of this paper (please number each point) 1. serious study 2. extensive experimental comparisons *11: Three weak points of this paper (please number each point) 1. limited novelty 2. limited improvement 3. unconvincing experimental validation *12: Detailed comments for the authors My concerns about the paper are as follows: 1. The experiments heavily compared with methods considering either network structure or node attribute alone, demonstrating improved accuracy. However, this is an unfair comparison and nothing to emphasize (can be included, but not a major contribution because you're not the first one to consider both data sources, and it's natural to do so). Those methods have their advantages, e.g., it may not be always the case that both data sources are available. 2. The paper claims scalability advantage over existing methods considering both data sources. Although it is indeed an advantage, but it's not critical because the community detection task is not time critical and such algorithms usually run offline. If you think scalability is critical, then motivate it and focus on it. Currently the paper lacks focus. It's a bit odd to include parallelization of NESNA. 3. The proposed algorithm CESNA extends BigCLAM and the novelty is limited. Overall, I'm not denying the contributions of the paper (I think it's definitely publishable) and improvements over existing approaches. I just feel they are not quite significant enough for this conference. 4. Some discussions are not very precise. e.g., the paper states that "clustering algorithms identify sets of objects whose attributes are similar, while ignoring relationships between objects. On the other hand, community detection algorithms aim to find communities based on the network structure." This may not be accurate. Community detection can be considered as an application of clustering, and clustering can definitely use both data sources. e.g., as in the following paper, where community identification is an application: Martin Ester, Rong Ge, Byron J. Gao, Zengjian Hu, Boaz Ben-Moshe: Joint Cluster Analysis of Attribute Data and Relationship Data: the Connected k-Center Problem. SDM 2006. 5. There are typos and grammar mistakes here and there. e.g., in the abstract, "focus on one one of these ..." For figures in the experiments, e.g., Fig. 4, try not to use only color to distinguish between the methods. Most printouts are in black and white. ======================================================== The review report from reviewer #2: *1: Is the paper relevant to ICDM? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [_] 4 (Innovative) [X] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [X] 4 (High) [_] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [X] 4 (Good) [_] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to ICDM users and practitioners? [X] 3 (Yes) [_] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [X] 2 (High) [_] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [_] 4 (Accept: a regular paper) [X] 3 (Weak Accept: a short paper) [_] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Summary of the paper's main contribution and impact This paper addresses the community detection problem in node-attributed networks. It extends a previous work on overlapping community detection [34], and proposes a generative model that exploits not only the structural information but also the node attributes. A scalable parallel algorithm is developed to learn the model. The proposed method is tested on 5 real world data sets and obtains promising results. *9: Justification of your recommendation Overlapping community detection is a practical problem with applications in social media and content sharing networks. This paper extends an existing solution that detects communities based on edge connections, and develops a scalable method that exploits auxiliary node attributes to boost detection accuracy. It is shown that the proposed method outperforms the state-of-the-art structure-based detection method, and can bring in extra insights into the detected communities based on the node attributes. Nonetheless, the novelty of this work is somewhat limited. The authors should also better position their work with regard to several related works on attributed graph clustering, which have been overlooked. *10: Three strong points of this paper (please number each point) 1. The paper tackles a practical problem with interesting applications. 2. The proposed method obtains promising empirical results. 3. The paper is clearly written. *11: Three weak points of this paper (please number each point) 1. The novelty of this work is somewhat limited. 2. Closely related works on attributed graph clustering have been overlooked. *12: Detailed comments for the authors The authors missed a number of techniques that have been developed for attributed graph clustering (see below for a few examples). These techniques also consider both structure and attribute information for node clustering. The authors are expected to better position their work against these alternatives. - Y. Zhou, H. Cheng, and J. X. Yu. Graph Clustering Based on Structural/Attribute Similarities, VLDB 2009. - Z. Xu, Y. Ke, Y. Wang, H. Cheng, and J. Cheng. A Model-Based Approach to Attributed Graph Clustering, SIGMOD 2012. The technical novelty of this paper seems limited. It mostly builds upon the framework of a previous work [34] and combines a bunch of standard techniques. The idea of exploiting node attributes to improve detection performance is also not new. In particular, the assumptions and proposed probabilistic model for node attributes are in spirit similar to that of Xu et al (2012). Minor points: - Last equation on page 3: P_uv(c) should be 1-P_uv(c). - In the paragraph discussing the choice of the number of communities, the notation of the number of communities should be C instead of K. - It would be more convincing to see the empirical comparisons by setting the number of communities to the ground truth than the one selected by the proposed method. The authors should also report the selected number and give a sense how close it is to the ground truth. - It is unclear where the error bars in the figures and the significance test come from. What is the cause of the randomness? Were the experiments run multiple times? ======================================================== The review report from reviewer #3: *1: Is the paper relevant to ICDM? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [_] 4 (Innovative) [X] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [X] 4 (High) [_] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [X] 4 (Good) [_] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to ICDM users and practitioners? [X] 3 (Yes) [_] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [X] 2 (High) [_] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [_] 4 (Accept: a regular paper) [X] 3 (Weak Accept: a short paper) [_] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Summary of the paper's main contribution and impact The paper proposes CESNA that combines both network structure ad node attributes to detect communities. This paper is well written and it is pleasure to read through it. The extensive experiments on networks with ground truth show that the algorithm has improvement over competitors. Technically, it is an extension of the BigCLAM by considering the attributes, but has some novelty. I think the work makes some contribution to this specific line of research. *9: Justification of your recommendation see below *10: Three strong points of this paper (please number each point) Strong: 1. Well written 2. Extensive experiments 3. Technically sounds. *11: Three weak points of this paper (please number each point) I did not find significant weakness. *12: Detailed comments for the authors Overall, I like this paper. One comment I have is: to demonstrate that the work really adds value to the line of overlapping community detection, it will be encouraged to compare with more other state-of-the-art algorithms (e.g., [33]) and to avoid bias in the metrics, it will be good to evaluate with other network specific measures(e.g., NMI).