Research On Structure, Evolution And Prediction Of Co-authorship Networks

In recent years, complex network theory has attracted much attention. The research has been conducted in many areas including sociology, computer science, biology, physics and so on. Researchers, in different areas and from different point of views, are studying complex networks. People call it as "21st century science." Collaboration network, which is a kind of social network, is a hot research field in complex network analysis. Co-authorship network is a kind of collaboration network that has been analyzed at early stage. In the era of big data, the vast amounts of data provide good support for relevant studies. The research on co-authorship networks, including how collaborations are formed, the structures and patterns of collaborations, the evolution mechanisms of collaborations, could help finding the patterns of scientific collaborations, guiding scientific research activities, optimizing the structure of collaborations, and thus promoting the development progress of science and society.This study focuses on the analysis and research of the co-authorship networks, mainly from the following four parts:1. The construction of co-authorship networks:In co-authorship networks, the authors are nodes, and the collaborations between authors are edges. Literatures are the data source of co-authorship network construction and the basis of co-authorship network research. In this paper, we have two kinds of data sources. Firstly, we propose a crowdsourcing method based on Scholarometer system to collect author, paper and discipline information. The data could facilitate the analysis of co-authorship networks, discipline evolution and author impact measures. Secondly, we collect publically available data from web. For author-related research based on literatures, author name ambiguity is a big challenge. Previous studies used features such as co-authors and titles to detect ambiguous names. However, the results could be wrong for interdisciplinary authors. In this paper, we propose features such as name variations and citations and topic consistency, and incorporate various features to detect author name ambiguity. Compared with Microsoft Academic Search, our method could detect ambiguous authors more effectively. The method has been integrated into Scholarometer system in order to eliminate ambiguity and reduce noise data.2. The structure of co-authorship networks:The characteristics of network structure are closely related to the dynamics of behaviors on networks. In this paper, the basic attributes and dynamic attributes of co-authorship networks in different disciplines are analyzed to reveal the general patterns. The structural characteristics could guide the design of the microscopic mechanisms for dynamical models. Besides the analysis of structural characteristics, this paper studies community structure in the networks, which is an important structure in complex networks. Previous studies on community detection always focus on the structure and ignore structure-related context information. In this paper, under the context of social tagging systems, we propose a generative probabilistic model based on authors’ collaborative tagging behavior and interests. The model incorporates the tagging behavior and topical context information, which could discover communities that the nodes in which are both closely connected and have similar interests. Compared with other methods, our model could detect more realistic communities and benefit applications such as collaborator recommendation.3. The evolution of co-authorship networks:Understanding the formation mechanism of networks is very important for predicting the future evolution of networks and simulating large scale networks. There are two kinds of network evolution models:network models based on structure or node attributes. There is no standard answer that which mechanism is better. In this paper, we aim to combine these two mechanisms to design the evolution model. The characteristics of the collaborations between authors are analyzed in order to find effective features that influence collaborations. Then, guided by the effective features, an agent-based model is proposed to simulate the evolution of co-authorship networks. The model is driven by events and interests, and produces a number of network attributes and patterns observed in real data. Finally, the model is applied to predict the future collaborations from observed data. Compared with several link prediction methods, the proposed method achieves higher accuracy, which demonstrates the rationality and effectiveness of the model.4. Discipline evolution from co-authorship network perspective:Uncovering the patterns of the birth, evolution and decline of disciplines has always been an endeavor of researchers. Most of the existing discipline evolutionary models only have qualitative conclusions, which are difficult to be verified in experiments. In this paper, we aim to propose an quantitive model to describe the evolution of disciplines. Discipline is an important attribute of authors, and authors could belong to one or more disciplines. The collaborations between authors represent the collaborations between disciplines and collaborations between authors from different disciplines could indicate the emergence of interdisciplines. Therefore, we propose an agent-based model to analyze the evolution of disciplines from the perspective of co-authorships. In the model, disciplines are defined as groups of authors and the evolution of co-authorship networks is evaluated by modularity. The merge and split of co-authorship networks suggesting the merge of disciplines and the birth of new disciplines. The model could explain the stylized facts of the relationships between disciplines, authors and papers very well, and reveal the patterns of discipline evolution.From the studies of the above four parts, we could have a better understanding about the structure, evolution and prediction of co-authorship networks. The studies provide a reasonable explanation for co-authorship network evolution and discipline evolution, and benefit for researchers to understand the structure and evolution of co-authorship networks and the evolution of disciplines.