Adaptive Similarity Function with Structural Features of Network Embedding for Missing Link Prediction

Link prediction is a fundamental problem of data science, which usually calls for unfolding the mechanisms that govern the micro-dynamics of networks. In this regard, using features obtained from network embedding for predicting links has drawn widespread attention. Although methods based on edge features or node similarity have been proposed to solve the link prediction problem, many technical challenges still exist due to the unique structural properties of networks, especially when the networks are sparse. From the graph mining perspective, we first give empirical evidence of the inconsistency between heuristic and learned edge features. Then, we propose a novel link prediction framework, AdaSim, by introducing an Adaptive Similarity function using features obtained from network embedding based on random walks. The node feature representations are obtained by optimizing a graph-based objective function. Instead of generating edge features using binary operators, we perform link prediction solely leveraging the node features of the network. We define a flexible similarity function with one tunable parameter, which serves as a penalty of the original similarity measure. The optimal value is learned through supervised learning and thus is adaptive to data distribution. To evaluate the performance of our proposed algorithm, we conduct extensive experiments on eleven disparate networks of the real world. Experimental results show that AdaSim achieves better performance than state-of-the-art algorithms and is robust to different sparsities of the networks.

An Effective Approach for Modular Community Detection in Bipartite Network Based on Integrating Rider with Harris Hawks Optimization Algorithms

The strenuous mining and arduous discovery of the concealed community structure in complex networks has received tremendous attention by the research community and is a trending domain in the multifaceted network as it not only reveals details about the hierarchical structure of multifaceted network but also assists in better understanding of the core functions of the network and subsequently information recommendation. The bipartite networks belong to the multifaceted network whose nodes can be divided into a dissimilar node-set so that no edges assist between the vertices. Even though the discovery of communities in one-mode network is briefly studied, community detection in bipartite networks is not studied. In this paper, we propose a novel Rider-Harris Hawks Optimization (RHHO) algorithm for community detection in a bipartite network through node similarity. The proposed RHHO is developed by the integration of the Rider Optimization (RO) algorithm with the Harris Hawks Optimization (HHO) algorithm. Moreover, a new evaluation metric, i.e., h-Tversky Index (h-TI), is also proposed for computing node similarity and fitness is newly devised considering modularity. The goal of modularity is to quantify the goodness of a specific division of network to evaluate the accuracy of the proposed community detection. The quantitative assessment of the proposed approach, as well as thorough comparative evaluation, was meticulously conducted in terms of fitness and modularity over the citation networks datasets (cit-HepPh and cit-HepTh) and bipartite network datasets (Movie Lens 100 K and American Revolution datasets). The performance was analyzed for 250 iterations of the simulation experiments. Experimental results have shown that the proposed method demonstrated a maximal fitness of 0.74353 and maximal modularity of 0.77433, outperforming the state-of-the-art approaches, including h-index-based link prediction, such as Multiagent Genetic Algorithm (MAGA), Genetic Algorithm (GA), Memetic Algorithm for Community Detection in Bipartite Networks (MATMCD-BN), and HHO.

An Experimental Evaluation of Similarity-Based and Embedding-Based Link Prediction Methods on Graphs

The task of inferring missing links or predicting future ones in a graph based on its current structure is referred to as link prediction. Link prediction methods that are based on pairwise node similarity are well-established approaches in the literature and show good prediction performance in many realworld graphs though they are heuristic. On the other hand, graph embedding approaches learn lowdimensional representation of nodes in graph and are capable of capturing inherent graph features, and thus support the subsequent link prediction task in graph. This paper studies a selection of methods from both categories on several benchmark (homogeneous) graphs with different properties from various domains. Beyond the intra and inter category comparison of the performances of the methods, our aim is also to uncover interesting connections between Graph Neural Network(GNN)- based methods and heuristic ones as a means to alleviate the black-box well-known limitation.

An Experimental Evaluation of Similarity-Based and Embedding-Based Link Prediction Methods on Graphs

The task of inferring missing links or predicting future ones in a graph based on its current structure is referred to as link prediction. Link prediction methods that are based on pairwise node similarity are well-established approaches in the literature and show good prediction performance in many realworld graphs though they are heuristic. On the other hand, graph embedding approaches learn lowdimensional representation of nodes in graph and are capable of capturing inherent graph features, and thus support the subsequent link prediction task in graph. This paper studies a selection of methods from both categories on several benchmark (homogeneous) graphs with different properties from various domains. Beyond the intra and inter category comparison of the performances of the methods, our aim is also to uncover interesting connections between Graph Neural Network(GNN)- based methods and heuristic ones as a means to alleviate the black-box well-known limitation.

Link prediction based on heterogeneous degree penalization with extending neighbors and clustering coefficient

Link prediction, aiming to find missing links in a current network or to predict some possible new links in a future network, is a challenging problem in complex networks. Many existing link prediction algorithms perform the task by optimizing the node similarity measures, and then determining the possibility of the link between any pair of similar nodes. In this paper, we propose a novel node similarity index named heterogeneous degree penalization (HDP), which incorporates the quasi-local structure information of extending neighborhood of each pair of nodes to be predicted and the clustering coefficient of their common neighbors. For specific networks with different statistical properties, we can achieve a good performance of link prediction through adjusting the penalty weights. The experiment results show that, comparing with the other existing approaches, the proposed method can remarkably improve the accuracy of link prediction.

Appraisal Study of Similarity-Based and Embedding-Based Link Prediction Methods on Graphs

The task of inferring missing links or predicting future ones in a graph based on its current structure is referred to as link prediction. Link prediction methods that are based on pairwise node similarity are well-established approaches in the literature and show good prediction performance in many real-world graphs though they are heuristic. On the other hand, graph embedding approaches learn low-dimensional representation of nodes in graph and are capable of capturing inherent graph features, and thus support the subsequent link prediction task in graph. This appraisal paper studies a selection of methods from both categories on several benchmark (homogeneous) graphs with different properties from various domains. Beyond the intra and inter category comparison of the performances of the methods our aim is also to uncover interesting connections between Graph Neural Network(GNN)-based methods and heuristic ones as a means to alleviate the black-box well-known limitation.

A New Method of Identifying Core Designers and Teams Based on the Importance and Similarity of Networks

In the process of product collaborative design, the association between designers can be described by a complex network. Exploring the importance of the nodes and the rules of information dissemination in such networks is of great significance for distinguishing its core designers and potential designer teams, as well as for accurate recommendations of collaborative design tasks. Based on the neighborhood similarity model, combined with the idea of network information propagation, and with the help of the ReLU function, this paper proposes a new method for judging the importance of nodes—LLSR. This method not only reflects the local connection characteristics of nodes but also considers the trust degree of network propagation, and the neighbor nodes’ information is used to modify the node value. Next, in order to explore potential teams, an LA-LPA algorithm based on node importance and node similarity was proposed. Before the iterative update, all nodes were randomly sorted to get an update sequence which was replaced by the node importance sequence. When there are multiple largest neighbor labels in the propagation process, the label with the highest similarity is selected for update. The experimental results in the related networks show that the LLSR algorithm can better identify the core nodes in the network, and the LA-LPA algorithm has greatly improved the stability of the original LPA algorithm and has stably mined potential teams in the network.

A Node Similarity and Community Link Strength-Based Community Discovery Algorithm

Community structure is one of the common characteristics of complex networks. In the practical work, we have noted that every node and its most similar node tend to be assigned to the same community and that two communities are often merged together if there exist relatively more edges between them. Inspired by these observations, we present a community-detection method named NSCLS in this paper. Firstly, we calculate the similarities between any node and its first- and second-order neighbors in a novel way and then extract the initial communities from the network by allocating every node and its most similar node to the same community. In this procedure, some nodes located at the community boundaries might be classified in the incorrect communities. To make a redemption, we adjust their community affiliations by reclassifying each of them into the community in which most of its neighbors have been. After that, there might exist relatively larger number of edges between some communities. Therefore, we consider to merge such communities to improve the quality of the final community structure further. To this end, we calculate the link strength between communities and merge some densely connected communities based on this index. We evaluate NSCLS on both some synthetic networks and some real-world networks and show that it can detect high-quality community structures from various networks, and its results are much better than the counterparts of comparison algorithms.