Co-authorship Networks in Additive Manufacturing Studies Based on Social Network Analysis

.


INTRODUCTION
Additive Manufacturing (AM), also known as 3D printing, has been undergoing tremendous growth in the recent years. As opposed to subtractive manufacturing methodologies, such as traditional machining, AM is a process of making objects from 3D model data by joining materials layer by layer, whether the material is plastic, metal or concrete. Its novelty stems from the layer-wise deposition principle offers a range of possibilities including multi-material structures and mesoscopic devices. AM technologies have been adopted by many industries for designing, development, rapid prototyping and manufacturing purposes because of its immense advantages related to manufacturing and production. Its impact is expected to go beyond market predictions across multiple industries in the future [1].
Scientific research is becoming increasingly difficult for individuals in era of knowledge economy. Cooperation has become a common phenomenon in scientific research, which usually leads to a scientific output of a greater quality than an individual [2]. As a consequence of this trend, there has been increasing academic interest in the phenomenon of co-authorship among scholars [3]. Co-authorship involves the participation of two or more authors in the production of a study. The results of scientific cooperation are usually co-authored published in academic journals, so it is an important way to analysis academic journals in order to summarize the characteristics of technology development in the field of AM. Social network analysis (SNA) provides a more effective analysis way for collaborative research paper [3,4].
A Social network involves relationships or interactions [5]. It analyses social structures and networks, where the nodes consist of scholars or actors, and the edges consist of the coauthorship or co-starship among these actors. While data crawling technology is a subsidiary premise for study on online social network, the complex and large-scale of World Wide Web bring a lot of problems for data collection, such as dynamic data is unavailable, and measurement frequency is confining. Reasonable reptile crawling strategy is an important guarantee for data analysis.
In this paper, the co-authorship networks on the field of AM are paid particular attention. A framework of data collection based on open source software is proposed. The characteristics of the overall network structure is analyzed and visualized. Based on the results of subnet analysis, the development process of AM collaboration community is discussed. So as to provide inspiration for knowledge exchange and further development of research in the field of AM. The paper presents the main conclusions and limitations of the study at last.

DATA COLLECTION ARCHITECTURE FOR ACADEMIC DATABASES
For this study, we collected data from the Web of Science of Thomson Reuters database. This database provides access to information from approximately 8,500 of the most prestigious, influential research journals in the world.
Since AM technology has developed rapidly in recent years, thousands of articles have published. It is infeasible to gathers the publication information manually. On the other hand, various academic databases limited the data access. For example, the BIBTEX entries exported is limited up to 500 once in the Web of Science database, which increases the difficulty of data acquisition. A framework based on open source software is proposed to collect data according to the need, since avoiding the reinvention of the proverbial wheel is a standard bit of received wisdom in software development circles [6].
The core in this framework is Scrapy [7]. It is based on the python language and uses an asynchronous network library to handle network communications, extract structured data. Scrapy has a clear architecture and contains a variety of middleware interfaces. Its flexible architecture meets a variety of needs, such as information processing and data mining. All that is needed is to customize given modules in order to crawl the web text as well as a variety of pictures.
The software architecture shown in Fig. 1 The RandomUserAgentMiddleware is tak Scrapy to prevent a crawler being identified and blocked by servers because of the default user agent or a generic one being frequently used. The Bibtex is selected as the exported file format, which correspondingly is exported to GraphML [8] file format. So the IGraph software package for complex network can analyze the characteristic of co networks in the field of AM. The RandomUserAgentMiddleware is taken in Scrapy to prevent a crawler being identified and blocked by servers because of the default useragent or a generic one being frequently used. The Bibtex is selected as the exported file format, which correspondingly is exported to file format. So the IGraph [9] software package for complex network can analyze the characteristic of co-authorship Downloader

Web of Knowledge
In this paper, ISI Web of Science in SCI-Expanded database was selected as a data source. In the process of development of AM, a variety of names have been proposed, such as rapid manufacturing (RM), rapid prototyping (RP), rapid prototype manufacturing (RPM), and layered manufacturing technology (LMT). So the retrieval expression was "TS = ((additive manufacturing) OR (3D printing) OR (rapid prototyping) OR (rapid prototyping manufacturing) OR (layered manufacturing technology))". The retrieval language was set to English, and the document type was set to Article. The Timespan was set from 2006 to 2016. A total of 8981 articles were retrieved.
We can see from Fig. 2

ADDITIVE MANUFACTURING CO-AUTHORSHIP NETWORKS
Data about 8981 articles, 34208 authors, and 102418 co-authorships was finally available for our analysis.
Since the more important scholars should have published more articles, scholars who published more than 4 articles is chosen as the research object of co-authorship networks. Totals 385 scholars with 560 times co-authored experiments meets this condition. The research collaboration network is represented through a graph. Fig. 3 shows the co-authorship network of AM community. Each node of the graph represents an author. The connection between the nodes indicates the collaboration relationship between scholars. The weight of a link indicates the number of publications that two scholars

Subnet Pattern Analysis
According to the definition of AM community, there are 57 co-author subnets except 58 isolated nodes. The largest connected component consists of 106 nodes. The disregarding weights is 0.00758, which indicates the overall network structure is relatively loose. The co-authorship in this AM field is too sparse to form a complete network connectivity, which manifests the need for more academic communication and cooperation in the development process of AM.
Different subnet have different characteristics, the subnets of overall network are classified and its network characteristics is analyzed, which contributes to further understanding the current state of development and future trends of AM field. The co-author subnets are classified by single-point mode, dual-core mode, bridge mode, core mode and grid mode [10]. The community type distribution in the field of AM is shown in Fig. 4.

Fig. 4. The community type distribution in the field of AM
Single-point mode indicates the author publishes an article by themselves.
Dual-core mode is the co-authorship between two authors. There are 25 subnets in dual-core mode. The two authors are usually at the same research institution in favor of exchanging views opinions timely. However, this mode lacks interagency communication, which is easy to imprison research methods, so it is not conducive to long-term development of the AM technology.
Bridge mode refers to a subnet in which one or two nodes are connected to two smaller subnets.
The key nodes in the bridge mode not only have the more research results, but also the number of co-authored articles with others in the subnets are the most. Since two subnets can only exchange information through the key nodes, this mode is extremely unfavorable to the flow of information. Once the key nodes damage, the entire bridge network will split into two smaller subnets.
Core mode refers to the network that has an important node, while the other nodes are connected with that node. Core Mode subnet is in the form of the center outward diffusion. Divergent subnet structure is unstable. The core nodes understand, master, and even destroy the network. If the core nodes once lost, exchange and cooperation within the network will be fully destroyed, which results in communications interruption, information paralysis. The entire network will cease to exist in the worst case.
Grid mode refer to a subnet in which any of nodes have a direct co-authorship or indirect relationship through another node. Losing any of nodes does not affect the entire network connectivity. This is an ideal information sharing structure. The authors in grid mode subnet have very good cooperation. The mode is mostly based on the same research institution or joint research projects. To attract new participator, the subnet members need to break its original coauthorship, and form a grid mode subnet again. The subnet will get better development in the way.

AM Technology Prospect
The co-authorship among researchers in the field of AM is sparse, which hints a large number of small subnets exists. The cooperation and exchange between small subnets are insufficiency. As an emerging field of research, AM community needs to take some time to accumulate in order to realize the effective communication and cooperation in the field of knowledge.
According to Fig. 4, the current co-authorship in the field of AM is diverse. The subnet that contains only two or three authors are widespread. AM field does not see a large and complex network of cooperation. While Grid mode is made up of interdisciplinary research team, whose subnet structure is conducive to complementary advantages between several small research team. High quality research results can be achieved by promoting interdisciplinary diffusion of knowledge.

Conclusions
Research and development of emerging technologies such as AM are knowledge production process. More breakthroughs in the field of technical problems and exploration in this process of innovation require collaboration among interdisciplinary teams. Collaborative research has increasingly become the development trend of academic research.
The present study is to analyze collaboration in the field of AM by using co-authorships in scientific journals as an indicator. A framework based on open source software is proposed to collect data in the AM co-authorship network. The results show that the framework is effective in large-scale data mining. The current coauthorship in the field of AM is diverse, a number of authors are still isolated from the main networks, although this situation might be caused by selected authors. This research shows that a large-scale network which contains 106 authors with a high centrality exists.

Future Works
In this paper, only the authors who published more than 4 articles were analyzed, which may affect the comprehensiveness. Moreover, the study only considered the current co-authorship mode of the subnet, and there is no analysis in the formation reasons of the subnet. That is next work we need to solve in our future research.