======= Review 1 ======= > *** Contributions: What are the major issues addressed in the paper? Do you consider them important? Comment on the novelty, creativity, impact, and technical depth in the paper. This paper presents a method to identify and classify spatial-temporal call data for mobile phones. The authors are attempting to provide a generic framework that can be used at different spatial and temporal scales. The authors present two similarity metrics, which while straight forward seem to be novel, especially in this context. The problem presented by the authors, that better analysis technique for the spatial and temporal distribution of call volumes across an urban area will improve access network design is important, especially given the reliance on mobile network in the developing world. The authors method also proves to be good at detecting outliers related to real world events. > *** Strengths: What are the major reasons to accept the paper? [Be brief.] Very strong technical analysis, lots of details. Very good results. > *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.] Motivation is slightly weak as the authors do not go into how their results will be useful towards network design. > *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors. The sections at the end (Results, C & D) are a bit wordy at times and can be hard to follow, but the technical work and figures are quite clearly presented. The impact of the outlier detection will make for very interesting future work, perhaps in the context of detecting social unrest as it begins. > *** Recommendation: Your overall rating (Please try giving as few borderlines as possible). A = (top 10% of reviewer's perception of all INFOCOM submissions) strong accept (5) ======= Review 2 ======= > *** Contributions: What are the major issues addressed in the paper? Do you consider them important? Comment on the novelty, creativity, impact, and technical depth in the paper. This paper proposes a complete solution how to classify the call profile in a systematic way. > *** Strengths: What are the major reasons to accept the paper? [Be brief.] The proposed solution can automatically classify the call profiles for a large amount of traffic data. The performance proposed solution is verified using real traffic traces. > *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.] Some design choices can be better explained. > *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors. The authors proposed a 2-staged classification scheme, which first builds the clusters through a small portion of learning data. Then it uses the clusters to classify the call profile of the rest of the data. Two similarity metrics are used, the traffic volume pattern similarity and the normalized traffic distribution pattern similarity. Although the authors claim that this two similarity metrics are orthogonal, the reasons are not 100% clear to me. I think the (the sum of all traffic) and (the distribution of all traffic) are orthogonal but (the volume of all traffic) and the (distribution of all traffic) are still correlated. Some further discussion on this point will be appreciated. Then two different index methods are then used to aggregate the dendrogram. If the authors could give some analytical performance characterization of their proposed scheme, the paper can be further improved. > *** Recommendation: Your overall rating (Please try giving as few borderlines as possible). B+ = (top 20% of reviewer's perception of all INFOCOM submissions, but not top 10%) weak accept (4) ======= Review 3 ======= > *** Contributions: What are the major issues addressed in the paper? Do you consider them important? Comment on the novelty, creativity, impact, and technical depth in the paper. Classifying call profiles is very useful to the access network optimizations. This paper gives a snapshot based call profiles classification method. Different with previous methods, this kind of classification is more fine-grained. The authors designed and verified the method based on a huge datasets which has 300 million calls over 5 months. > *** Strengths: What are the major reasons to accept the paper? [Be brief.] The snapshots based method can reveal more information corresponded to the whole data aggregation methods. A large amount of measurement data from practical is considered. > *** Weaknesses: What are the major reasons NOT to accept the paper? [Be brief.] The results are not convinced. It is lack of quantitative analysis to explain this method is better than other. The algorithm is intuitive. And the effectiveness should be proved. > *** Detailed Comments: Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors. Classifying call profiles is very useful to the access network optimizations. This paper gives a snapshot based call profiles classification method. Different with previous methods, this kind of classification is more fine-grained. The authors designed and verified the method based on a huge datasets which has 300 million calls over 5 months. In general speaking, this is a good paper that gives a very interesting study. It however needs significant work on the paper presentation as the current paper is really hard to follow. Some key questions are not well addressed in the paper. For example, we are not very clear how to use this classification result. In the paper, authors referred that there were 30 candidates stopping rules and simply select foure of them. To our understanding, these rules will play critical role for the performance. A nature question is, why we select these four rules? Is there any design philosophies in behind, or the authors randomly select these rules? What are the impacts of these rules? For the similarity, authors adopt the PPMCC, and why we have such selection? As there are many other similarity definiations such as the original CC, cosine, and etc, do we have other design options? The answered questions are much more than answered ones. Considering the large amount of traces collected in the study, we suggest an accept if there is room. > *** Recommendation: Your overall rating (Please try giving as few borderlines as possible). B = (top 30% of reviewer's perception of all INFOCOM submissions, but not top 20%) - accept if there is room (3)