Sandipan Sikdar

Computer science fields as ground-truth communities: their impact, rise and fall

Study of community in time-varying graphs has been limited to its detection and identification across time. However, presence of time provides us with the opportunity to analyze the interaction patterns of the communities, understand how each individual community grows/shrinks, becomes important over time. This paper, for the first time, systematically studies the temporal interaction patterns of communities using a large scale citation network (directed and unweighted) of computer science. Each individual community in a citation network is naturally defined by a research field - i.e., acting as ground-truth - and their interactions through citations in real time can unfold the landscape of dynamic research trends in the computer science domain over the last fifty years. These interactions are quantified in terms of a metric called inwardness that captures the effect of local citations to express the degree of authoritativeness of a community (research field) at a particular time instance. Several arguments to unveil the reasons behind the temporal changes of inwardness of different communities are put forward using exhaustive statistical analysis. The measurements (importance of field) are compared with the project funding statistics of NSF and it is found that the two are in sync. We believe that this measurement study with a large real-world data is an important initial step towards understanding the dynamics of cluster-interactions in a temporal environment. Note that this paper, for the first time, systematically outlines a new avenue of research that one can practice post community detection.

Citation interactions among computer science fields: a quantitative route to the rise and fall of scientific research

In this work, we propose for the first time a suite of metrics that can be used to perform post-hoc analysis of the temporal communities of a large-scale citation network of the computer science domain. Each community refers to a particular research field in this network, and therefore, they act as natural sub-groupings of this network (i.e., ground-truths). The interactions between these ground-truth communities through citations over the real time naturally unfold the evolutionary landscape of the dynamic research trends in computer science. These interactions are quantified in terms of a metric called inwardness that captures the effect of local citations to express the degree of authoritativeness of a community (research field) at a particular time instance. In particular, we quantify the impact of a field, the influence imparted by one field on the other, the distribution of the “star” papers and authors, the degree of collaboration and seminal publications to characterize such research trends. In addition, we tear the data into three subparts representing the continents of North America, Europe and the rest of the world, and analyze how each of them influences one another as well as the global dynamics. We point to how the results of our analysis correlate with the project funding decisions made by agencies like NSF. We believe that this measurement study with a large real-world data is an important initial step towards understanding the dynamics of cluster-interactions in a temporal environment. Note that this paper, for the first time, systematically outlines a new avenue of research that one can practice post community detection.

Time series analysis of temporal networks

AbstractA common but an important feature of all real-world networks is that they are temporal in nature, i.e., the network structure changes over time. Due to this dynamic nature, it becomes difficult to propose suitable growth models that can explain the various important characteristic properties of these networks. In fact, in many application oriented studies only knowing these properties is sufficient. For instance, if one wishes to launch a targeted attack on a network, this can be done even without the knowledge of the full network structure; rather an estimate of some of the properties is sufficient enough to launch the attack. We, in this paper show that even if the network structure at a future time point is not available one can still manage to estimate its properties. We propose a novel method to map a temporal network to a set of time series instances, analyze them and using a standard forecast model of time series, try to predict the properties of a temporal network at a later time instance. To our aim, we consider eight properties such as number of active nodes, average degree, clustering coefficient etc. and apply our prediction framework on them. We mainly focus on the temporal network of human face-to-face contacts and observe that it represents a stochastic process with memory that can be modeled as Auto-Regressive-Integrated-Moving-Average (ARIMA). We use cross validation techniques to find the percentage accuracy of our predictions. An important observation is that the frequency domain properties of the time series obtained from spectrogram analysis could be used to refine the prediction framework by identifying beforehand the cases where the error in prediction is likely to be high. This leads to an improvement of 7.96% (for error level ≤20%) in prediction accuracy on an average across all datasets. As an application we show how such prediction scheme can be used to launch targeted attacks on temporal networks.

Identifying and Characterizing Sleeping Beauties on YouTube

The generally accepted notion about popularity dynamics of user generated contents (e.g., tweets, videos) is that such contents attain their peak popularity within first few days and then gradually fade into oblivion. However, analyzing more than 350K videos on YouTube, we find that more than 10% of them obtain their peak popularity after at least one year from being uploaded. We term such videos as Sleeping Beauties and observe that these videos engage users more compared to other videos on YouTube. We further observe that sleeping beauties can retain their popularity to a greater extent following their peak popularity compared to other videos. We believe that recognizing such videos will not only benefit the advertisers, but also the designers of recommendation systems who seek to maximize user satisfaction. Through this interactive poster, we bring the presence and characteristics of sleeping beauties in front of the research community.

On the effectiveness of the scientific peer-review system: a case study of the Journal of High Energy Physics

The importance and the need for the peer-review system is highly debated in the academic community, and recently there has been a growing consensus to completely get rid of it. This is one of the steps in the publication pipeline that usually requires the publishing house to invest a significant portion of their budget in order to ensure quality editing and reviewing of the submissions received. Therefore, a very pertinent question is if at all such investments are worth making. To answer this question, in this paper, we perform a rigorous measurement study on a massive dataset (29 k papers with 70 k distinct review reports) to unfold the detailed characteristics of the peer-review process considering the three most important entities of this process—(i) the paper (ii) the authors and (iii) the referees and thereby identify different factors related to these three entities which can be leveraged to predict the long-term impact of a submitted paper. These features when plugged into a regression model achieve a high

Threshold-based epidemic dynamics in systems with memory

R^2

Influence of reviewer interaction network on long-term citations: a case study of the scientific peer-review system of the journal of high energy physics

R 2 of 0.85 and RMSE of 0.39. Analysis of feature importance indicates that reviewer- and author-related features are most indicative of long-term impact of a paper. We believe that our framework could definitely be utilized in assisting editors to decide the fate of a paper.

Anomalies in the Peer-review System: A Case Study of the Journal of High Energy Physics

Vertices with high betweenness and closeness centrality represent influential entities in a network. An important problem for time varying networks is to know a-priori, using minimal computation, whether the influential vertices of the current time step will retain their high centrality, in the future time steps, as the network evolves. In this paper, based on empirical evidences from several large real world time varying networks, we discover a certain class of networks where the highly central vertices are part of the innermost core of the network and this property is maintained over time. As a key contribution of this work, we propose novel heuristics to identify these networks in an optimal fashion and also develop a two-step algorithm for predicting high centrality vertices. Consequently, we show for the first time that for such networks, expensive shortest path computations in each time step as the network changes can be completely avoided; instead we can use time series models (e.g., ARIMA as used here) to predict the overlap between the high centrality vertices in the current time step to the ones in the future time steps. Moreover, once the new network is available in time, we can find the high centrality vertices in the top core simply based on their high degree. To measure the effectiveness of our framework, we perform prediction task on a large set of diverse time-varying networks. We obtain F1-scores as high as 0.81 and 0.72 in predicting the top m closeness and betweenness centrality vertices respectively for real networks where the highly central vertices mostly reside in the innermost core. For synthetic networks that conform to this property we achieve F1-scores of 0.94 and 0.92 for closeness and betweenness respectively. We validate our results by showing that the practical effects of our predicted vertices match the effects of the actual high centrality vertices. Finally, we also provide a formal sketch demonstrating why our method works.