Learning Advisor-Advisee Relationship from Multiplex Network Structure

Multiplex network community detection algorithm based on motif awareness

Multiplex networks are representations of multilayer interconnected complex networks where the nodes are the same at every layer. They turn out to be good abstractions of the intricate connectivity of multimodal transportation networks, among other types of complex systems. One of the most important critical phenomena arising in such networks is the emergence of congestion in transportation flows. Here, we prove analytically that the structure of multiplex networks can induce congestion for flows that otherwise would be decongested if the individual layers were not interconnected. We provide explicit equations for the onset of congestion and approximations that allow us to compute this onset from individual descriptors of the individual layers. The observed cooperative phenomenon is reminiscent of Braess' paradox in which adding extra capacity to a network when the moving entities selfishly choose their route can in some cases reduce overall performance. Similarly, in the multiplex structure, the efficiency in transportation can unbalance the transportation loads resulting in unexpected congestion.

High-centrality nodes have disproportionate influence on the behavior of a network; therefore controlling such nodes can efficiently steer the system to a desired state. Existing multiplex centrality measures typically rank nodes assuming the layers are qualitatively similar. Many real systems, however, are comprised of networks heterogeneous in nature, for example, social networks may have both agnostic and affiliative layers. Here, we use rank aggregation methods to identify intervention targets in multiplex networks when the structure, the dynamics, and our intervention goals are qualitatively different for each layer. Our approach is to rank the nodes separately in each layer considering their different function and desired outcome, and then we use Borda count or Kemeny aggregation to identify a consensus ranking—top nodes in the consensus ranking are expected to effectively balance the competing goals simultaneously among all layers. To demonstrate the effectiveness of consensus ranking, we apply our method to a degree-based node removal procedure such that we aim to destroy the largest component in some layers, while maintaining large-scale connectivity in others. For any multi-objective intervention, optimal targets only exist in the Pareto-sense; we, therefore, use a weighted generalization of consensus ranking to investigate the trade-off between the competing objectives. We use a collection of model and real networks to systematically investigate how this trade-off is affected by multiplex network structure. We use the copula representation of the multiplex centrality distributions to generate model multiplex networks with given rank correlations. This allows us to separately manipulate the marginal centrality distribution of each layer and the interdependence between the layers, and to investigate the role of the two using both analytical and numerical methods.

Multiplex networks describe a large number of systems ranging from social networks to the brain. These multilayer structure encode information in their structure. This information can be extracted by measuring the correlations present in the multiplex networks structure, such as the overlap of the links in different layers. Many multiplex networks are also weighted, and the weights of the links can be strongly correlated with the structural properties of the multiplex network. For example, in multiplex network formed by the citation and collaboration networks between PRE scientists it was found that the statistical properties of citations to coauthors differ from the one of citations to noncoauthors, i.e., the weights depend on the overlap of the links. Here we present a theoretical framework for modeling multiplex weighted networks with different types of correlations between weights and overlap. To this end, we use the framework of canonical network ensembles, and the recently introduced concept of multilinks, showing that null models of a large variety of network structures can be constructed in this way. In order to provide a concrete example of how this framework apply to real data we consider a multiplex constructed from gene expression data of healthy and cancer tissues.

The overwhelming usage of social and collaboration networks provides the opportunity to analyze evolution of relationships among individuals, like celebrities or co-authors. Discovering such phenomenon in large complex networks is non-trivial due to their large sizes. In this situation, the aggregation functions used in OLAP, are useful to analyze the summarized data. OLAP has successfully proven its worth on multidimensional or complex networks. However, existing aggregations in the current OLAP systems do not produce versatile results in case of social and collaboration networks. This happens because said type of networks have structural connectivity/links among nodes, which cannot be considered by OLAP during its execution. In this situation, a useful discovery in terms of identifying pairs of nodes whose relationships is emerging in recent time, is missed. Such discovery of pairs of nodes is important for various applications such as targeted marketing, future joint partnerships and predicting future correspondence to name a few. In this study, we call such pairs as Rising_Pairs and propose an aggregation function for performing OLAP on network data whose historical information is maintained over a period of times. Using structural information, Rising_Pairs, our proposed aggregation function, discovers the strongly coupled pairs in a network data by emphasizing their recent interactions and attribute similarities. In this way, useful information related to strongly coupled pairs in a network is identified. To verify the effectiveness of our proposal, we implemented it on various types of real-world networks like Facebook, Digital Bibliography and Library Project (DBLP) and Global Positioning System (GPS) trajectory datasets and observed interesting patterns.

Learning low-dimensional representations of graphs has facilitated the use of traditional machine learning techniques to solving classic network analysis tasks such as link prediction, node classification, community detection, etc. However, to date, the vast majority of these learning tasks are focused on traditional single-layer/unimodal networks and largely ignore the case of multiplex networks. A multiplex network is a suitable structure to model multi-dimensional real-world complex systems. It consists of multiple layers where each layer represents a different relationship among the network nodes. In this work, we propose MUNEM, a novel approach for learning a low-dimensional representation of a multiplex network using a triplet loss objective function. In our approach, we preserve the global structure of each layer, while at the same time fusing knowledge among different layers during the learning process. We evaluate the effectiveness of our proposed method by testing and comparing on real-world multiplex networks from different domains, such as collaboration network, protein-protein interaction network, online social network. Finally, in order to deliberately examine the effect of our model’s parameters we conduct extensive experiments on synthetic multiplex networks.

Bug triaging is an essential activity of defect repair, which is closely related to the cost of software maintenance. Researchers have proposed automatic bug triaging approaches to recommend bug fixers more efficiently and accurately. In addition to text features, most of the previous studies focused on single-layer bug tossing (or reassignment) graphs, but they ignored the multiplex (or multi-layer) network characteristics of human cooperative behavior. In this study, we build a collaborative multiplex network composed of a tossing graph and an e-mail communication graph in the bug triaging process. By integrating the idea of network embedding and multiplex network measures, we propose a new strategy of random walks. Moreover, we present a bug fixer prediction model that takes structure and text features as inputs. Experimental results on two large-scale open-source projects show that the proposed method outperforms the selected baseline approaches in terms of commonly-used evaluation metrics.

Complex networks recently opened new ways for investigating how language use is influenced by the mental representation of word similarities. This work adopts the framework of multiplex lexical networks for investigating lexical retrieval from memory. The focus is on priming, i.e., exposure to a given stimulus facilitating or inhibiting retrieval of a given lexical item. Supported by recent findings of network distance influencing lexical retrieval, the multiplex network approach tests how the layout of hundreds of thousands of word‐word similarities in the mental lexicon can lead to priming effects on multiple combined semantic and phonological levels. Results provide quantitative evidence that phonological priming effects are encoded directly in the multiplex structure of the mental representation of words sharing phonemes either in their onsets (cohort priming) or at their ends (rhyme priming). By comparison with randomised null models, both cohort and rhyming effects are found to be emerging properties of the mental lexicon arising from its multiplexity. These priming effects are absent on individual layers but become prominent on the combined multiplex structure. The emergence of priming effects is displayed both when only semantic layers are considered, an approximated representation of the so‐called semantic memory, and when semantics is enriched with phonological similarities, an approximated representation of the lexical‐auditory nature of the mental lexicon. Multiplex lexical networks can account for connections between semantic and phonological information in the mental lexicon and hence represent a promising modelling route for shedding light on the interplay between multiple aspects of language and human cognition in synergy with experimental psycholinguistic data.

Dynamics of a multiplex neural network with delayed couplings

Diversity in backgrounds, ideas, and beliefs plays an important role in the scientific research community yet researchers belonging to minority groups are often discriminated against in academic collaborations. Here I replicate game-theoretic models originally described in a 2018 publication by Hannah Rubin and Cailin O'Connor in an attempt to reproduce their findings. Findings from these replicated models followed the same trends reported by Rubin and O'Connor including an increased likelihood of discrimination associated with smaller minority group sizes as well as a decrease in researchers working with out-group partners as a result of discriminatory norms in collaborative research networks. I then build on these base models and their findings to propose future model extensions that will provide insight into both the short-term and long-term impacts of discrimination faced by minority researchers within and beyond collaboration networks.

Scholars in academia are involved in various social relationships such as advisor-advisee relationships. The analysis of such relationship can provide invaluable information for understanding the interactions among scholars as well as providing many researcher-specific applications such as advisor recommendation and academic rising star identification. However, in most cases, high quality advisor-advisee relationship dataset is unavailable. To address this problem, we propose Shifu, a deep-learning-based advisor-advisee relationship identification method which takes into account both the local properties and network characteristics. In particular, we explore how to crawl advisor-advisee pairs from PhDtree project and extract their publication information by matching them with DBLP dataset as the experimental dataset. To the best of our knowledge, no prior effort has been made to address the scientific collaboration network features for relationship identification by exploiting deep learning. Our experiments demonstrate that the proposed method outperforms other state-of-the-art machine learning methods in precision (94%). Furthermore, we apply Shifu to the entire DBLP dataset and obtain a large-scale advisor-advisee relationship dataset.

Collaboration is increasingly becoming a vital factor in an academic network, which can bring lots of benefits for scholars. Ubiquitous intelligence also provides an effective way for scholars to find collaborators. However, due to the large-scale of scholarly big data, there is a lot of information hard to capture in networks and we need to dig out valid information from collaboration networks. It is a valuable and urgent task to find appropriate collaborators for scholars. To address these problems, we hypothesize that fusing topic model and structure information could improve the performance of recommendations. In this paper, we propose a collaborator recommendation system, named TNERec (Topic-aware Network Embedding for scientific collaborator Recommendation), learning representations from scholars' research interests and network structure. TNERec first extracts scholars' research interests based on topic model and then learns vectors of scholars with network embedding. Finally, top-k recommendation list is generated based on the scholar vectors. Experimental results on a real-world dataset show the effectiveness of the proposed framework compared with state-of-the-art collaboration recommendation baselines.

Online users are typically active on multiple social media networks (SMNs), which constitute a multiplex social network. With improvements in cybersecurity awareness, users increasingly choose different usernames and provide different profiles on different SMNs. Thus, it is becoming increasingly challenging to determine whether given accounts on different SMNs belong to the same user; this can be expressed as an interlayer link prediction problem in a multiplex network. To address the challenge of predicting interlayer links, feature or structure information is leveraged. Existing methods that use network embedding techniques to address this problem focus on learning a mapping function to unify all nodes into a common latent representation space for prediction; positional relationships between unmatched nodes and their common matched neighbors (CMNs) are not utilized. Furthermore, the layers are often modeled as unweighted graphs, ignoring the strengths of the relationships between nodes. To address these limitations, we propose a framework based on multiple types of consistency between embedding vectors (MulCEVs). In MulCEV, the traditional embedding-based method is applied to obtain the degree of consistency between the vectors representing the unmatched nodes, and a proposed distance consistency index based on the positions of nodes in each latent space provides additional clues for prediction. By associating these two types of consistency, the effective information in the latent spaces is fully utilized. In addition, MulCEV models the layers as weighted graphs to obtain representation. In this way, the higher the strength of the relationship between nodes, the more similar their embedding vectors in the latent representation space will be. The results of our experiments on several real-world and synthetic datasets demonstrate that the proposed MulCEV framework markedly outperforms current embedding-based methods, especially when the number of training iterations is small.

Multiplex networks have been widely used to describe the multi-type connections of entities in the real world. However, researches for multiplex networks visualization unilaterally focus on the presentation of topological structure, lacking of specific high-level information presentation for quantitative comparison of interlayer structure. Users cannot participate in the exploration and freely choose the layers (or sub-graphs, regions, etc.) of interest for structural comparison. Contraposing the layer-edge patterns visual analysis tasks of multiplex networks, this paper puts forward a novel solution for exploration and analysis that tightly couples topological structure and high-level patterns. It mainly contains a multi-force directed model to realize the balanced layout of nodes in multi-layer topology, as well as two kinds of high-level patterns of which the visual representations are, respectively, designed by a familiar metaphor—that is, the similar pattern representation based on the area-proportional Venn diagrams and the interaction pattern representation based on the directed arrows. Furthermore, views association is implemented through underlying data sharing and multiple interactions which can be used to gain insights through the creation of selections of interest and produce high-level infographic-style overviews simultaneously. The experiments on real-world data demonstrate the support of the proposed method for layer-edge patterns analysis tasks in multiplex networks and the effectiveness for analyzing the multi-layer structure of multiplex networks.

Collaboration networks are a ubiquitous way to characterize the interactions between people. In this paper, we consider the problem of inferring social relations in collaboration networks, such as the fields that researchers collaborate in, or the categories of projects that Github users work on together. Social relation inference can be formalized as a multi-label classification problem on graph edges, but many popular algorithms for semi-supervised learning on graphs only operate on the nodes of a graph. To bridge this gap, we propose a principled method which leverages the natural homophily present in collaboration networks. First, observing that the fields of collaboration for two people are usually at the intersection of their interests, we transform an edge labeling into node labels. Second, we use a label propagation algorithm to propagate node labels in the entire graph. Once the label distribution for all nodes has been obtained, we can easily infer the label distribution for all edges. Experiments on two large-scale collaboration networks demonstrate that our method outperforms the state-of-the-art methods for social relation inference by a large margin, in addition to running several orders of magnitude faster.