Development patterns of scientific communities in technology enhanced learning.
Technology Enhanced Learning (TEL) is an emerging research area in computer science. Similar to other sub-disciplines of computer science, conferences have a dominant role to play in the communication of TEL research. According to Microsoft Academic Search, there are 58 conferences in contrast to 18 journals in the computer education category as of April, 2012. Such domination raises questions regarding an understanding of the communities of conferences and their development patterns, in order to have an overview of the current research work in the TEL area. For researchers, understanding the community means getting to know the research environment, which leads to self-adaptation and, hopefully, improvement in the field. For conference organizers and stakeholders, an overview of their communities is important for maintaining, cultivating and promoting conferences.
The application of Social Network Analysis (SNA) in the field of digital library research is very promising in terms of knowledge discovery (Pham & Klamma, 2010; Pham, Klamma & Jarke, 2011). In particular, the structure of scientific collaboration can be researched in great detail using SNA associated with two distinct data sets: the co-authorship graph and the citation graph. The co-authorship graph reveals the contribution structures of a scientific community by disclosing who has collaborated with whom in terms of co-authoring papers. The citation graph discloses the influencing areas, conferences, and journals of a conference in terms of cited papers. Together, the two graphs allow a detailed analysis of the knowledge structure and flows within a particular scientific community, but also an analysis of the knowledge flows between adjacent scientific communities.
SNA has been proposed as a means of studying TEL communities. Kienle and Wessner (2005; 2006) as well as Hoadley (2005), studied the Computer Support for Collaborative Learning (CSCL) community using statistical analysis and the visualization of citation and collaboration data of several CSCL conferences, including also the program committee members and geographical data. They analyzed the development of the community and identified key members using simple statistics such as the number of participants over time, the proportion of new and old members, as well as the geographical distribution of the members. Similarly, Ochoa, Mendez and Duval (2009) analyzed the ED-MEDIA (World Conference on Educational Media and Technology) community. Besides statistical measures, they also used SNA metrics such as betweenness centrality to rank authors. Another work on the TEL community specific to a single conference series was the analysis of EC-TEL (European Conference on Technology Enhanced Learning) by Reinhardt, Meier, Drachsler and Sloep (2011).
SNA has also been applied in knowledge mapping research within scientometrics in order to understand the organization of scientific knowledge of all sciences as well as of a single discipline. Morris and McCain (1998) explored the interdisciplinary nature of medical informatics and its internal structure using inter-citation and co-citation analysis. A combination of Science Citation Index (SCI) and Social Sciences Citation Index (SSCI) data was used in this study. McCain (1998) performed the co-citation analysis for journals in the field of neural network research. Cluster analysis, principal component analysis and multidimensional scaling (MDS) maps were used to identify the main research areas. Regarding the computer science discipline, Ding, Chowdhury and Foo (2000) studied the relationship between journals in the area of information retrieval using the same techniques. Based on the SciSearch database, Tsay, Xu and Wu (2003) mapped semiconductor literature using co-citation analysis. The data sets used in these studies were rather small, ranging from tens to several hundred journals. In more recent work, Boyack, Borner and Klavans (2007) mapped the structure and evolution of chemistry research over a 30-year time frame. Based on a general map generated from the combined Science Citation Index Expanded (SCIE) and SSCI from 2002, the authors assigned journals to clusters using inter-citation counts. Journals were assigned to the chemistry domains using Journal Citation Ranking (JCR) categories. Then, maps of chemistry research during different time periods and at different domain levels were generated to expose the changes that have taken place over the 30 years of the development of chemistry research.
In this article, we are concerned with a different aspect of communities of TEL conferences, that is the dynamics and patterns of community development. Unlike journals, conferences facilitate communication between participants through face-to-face meetings, academic presentations and dissemination activities. A systematic comparison of the key features of scientific community shows that, depending on the duration of existence, different conferences exhibit different development patterns. Therefore, communication patterns and community building processes are of special interest. Through an analysis of community development patterns, the main purpose of this paper is to raise the awareness of conference organizers and stakeholders with regard to these development dynamics. Consequently, another aim is to provide information that facilitates the identification of strong and weak indicators within their community, and to provide hints for improvement.
Background and methodology
Conference community development model
We have proposed a model to explain the development pattern, as well as the collaborative and citation behavior in conferences and journals in Pham, Klamma and Jarke (2011); the model is displayed in Figure 1. We have shown in this previous study that this model can be used to describe and explain the community building process of many conferences in different areas of computer science in terms of co-authorship and citation networks.
The co-authorship network of a conference series consists of authors as nodes. There is an edge between two authors if they have co-authored at least one paper published in a conference event in that series (note that we use "event" to refer to one specific conference in a conference series, e.g., the ICALT 2011 event in the ICALT conference series). In the co-authorship network there are initially few connections between authors (born phase). After some events, author groups become apparent in the network (bonding phase), which are--in the best case--gradually integrated through publications that involve authors from more than one group (emergence phase). Over time, a conference series then typically forms a network topology that features a strongly connected core group of authors that is connected to other smaller groups (focused topology). Alternatively, the co-authorship can develop into an interdisciplinary topology where several groups are connected via some gatekeepers, but where there is no core group. It can also develop toward a hierarchical topology which exposes some "super gatekeepers" who connect a hierarchy of groups.
The citation network of a conference series is a graph whose nodes represent papers which were cited by papers published by that conference series. A directed edge from a citing paper to a cited paper represents a citation. Essentially, the citation network represents the body of literature cited by papers of a particular conference series. For the citation network, the development process may be different. When the conference series is focused at the beginning, its papers tend to cite a body of fundamental literature in the field, and the citation network should be very dense and have a large connected component (focused topology). When the conference series has an interdisciplinary nature, the citation network will contain several components and the connections between these components may either be found or they may not. If the core topic of the series is still developing, the citation network may remain in the bonding or emergence phases.
[FIGURE 1 OMITTED]
As an instantiation of the development model, Figure 2 illustrates the development of the ICALT co-authorship network since its first event in 2001, in three-year intervals. Each snapshot shows a network composed of authors (nodes) and their accumulated co-authorship relationships (edges). At the inaugural event in 2001 (born phase; note that ICALT actually started off as the International Workshop on Advanced Learning Technologies, IWALT, at Massey University in New Zealand in the year 2000; however the first DBLP-indexed conference event in the ICALT series took place in 2001) we see many "isolated" nodes in the network representing authors who had one or more papers published without any co-authors. Small groups are formed by authors who have co-authored one or more papers with at least one co-author. Three years later, at ICALT 2004, we witness the existence of some larger co-author groups, and also some bonding among author groups. In 2007, some larger author groups are already clearly discernible, and the network also starts to form a clearly visible large component of core authors in the emergence phase. By 2010, the largest component is beginning to actually deserve the label "giant component" and we see that many members of the giant component have co-authorship ties to other authors and author groups on the periphery of the giant component. Although it is evident at the bottom of each network snapshot that the network includes a large pool of unconnected authors and author groups, the ICALT author network tends toward developing a focused topology.
[FIGURE 2 OMITTED]
Time series analysis methodology
To qualitatively characterize the development process of the TEL community in relation to this development model, we applied a time series analysis on the networks to reveal the following social-network parameters over time: densification law (Leskovec, Kleinberg & Faloutsos, 2007), clustering coefficient, maximum betweenness, largest connected component, diameter, and average path length (Wasserman & Faust, 1995). These parameters, which are explained in the following paragraphs, enable us to explain the community building process in Figure 1. To interpret the shape of the community, one needs to use a combination of all of these parameters. Formally, given the network [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where V is the set of vertices or nodes, and E is the set of edges, the above network metrics are defined as follows:
Densification Jaw: Leskovec, Kleinberg and Faloutsos (2007) discovered that complex networks densify over time, with the number of edges growing super-linearly with the number of nodes, meaning that the average degree (i.e., number of edges) of the nodes is increasing. In fact, the densification follows a power-law pattern: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and n(t) are the number of edges and nodes at time t, respectively, and [varies] is an exponent that lies between 1 and 2 [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] corresponds to a constant average degree over time, while [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] corresponds to a very dense graph where, on average, each node has edges to a constant fraction of all nodes). We use exponent [varies] to differentiate the "speed" by which networks are densified.
Clustering coefficient measures the probability that two nodes are connected if they already have a common neighbor:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Intuitively, during the first phase of development, the clustering coefficient of the network is low, since nodes are unconnected with each other. In the second phase, the clustering coefficient tends to increase very quickly as nodes are clustered into very dense, yet unconnected components. When the unconnected components subsequently start to connect with each other, the clustering coefficient drops and stays relatively stable after some time.
Betweenness measures the extent to which a particular node lies between the other nodes in the network:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where B(u) is the betweenness of node u, [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is the number of shortest paths between nodes t and / that pass through u, and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is the number of shortest paths between nodes i and /. Nodes with high betweenness have more power to control the information flow in the network, and are normally the gatekeepers who connect several dense groups. For the overall network, the maximum betweenness of all authors is therefore a good indicator of whether there are strong gatekeepers within the network. During the first two phases of the development process, the maximum betweenness is very low, since the nodes are either completely unconnected or clustered in very dense, yet unconnected groups (i.e., there are no controllers in the network). Maximum betweenness increases when more components become connected (emergence stage) and continues to increase when the network develops toward a hierarchical or interdisciplinary topology. However, maximum betweenness will achieve a stable value when the network is at focused stage.
Largest connected component (or giant component) measures the fraction of nodes that are connected with each other in the largest sub-network. As observed in Figure 1, this fraction is small in the first two phases, and gradually increases as the network develops and authors from different sub-networks connect with each other. It achieves a stable state when the fraction of nodes that connect to the largest component is equal to the fraction of new nodes that stay unconnected from the largest component.
Diameter is the length of the greatest geodesic distance (i.e., the length of the longest shortest path) between any two nodes. Intuitively, in the beginning, the diameter is small, and then it increases. After some time, the diameter starts to shrink as new edges between existing nodes continue to be added. Note that the shrinking of the diameter is not caused by the emergence of the giant component (Leskovec, Kleinberg & Faloutsos, 2007). However, in our model, if the network develops toward a tree-like topology (hierarchical stage), the diameter will be larger than in the focused and interdisciplinary topologies, respectively.
Average path length is the average length of all the shortest paths in the network. Clearly, during the first two phases, the average path length is small and increases when the network grows. Although communities of conferences and journals are not random networks--and the average path length should therefore be rather small (around six)--there is a slight difference between focused, interdisciplinary and hierarchical topologies. In general, the average path length of a hierarchical network is larger than that of the other two topologies, which gives us more evidence to differentiate these topologies.
In summary, for the co-authorship network, the emergence of the giant component (largest connected component) indicates the cohesiveness of collaboration within the community, while the betweenness shows the existence of the gatekeepers and their importance. The clustering coefficient measures the extent to which the community is clustered into sub-communities. Other parameters such as diameter and average shortest path length, show whether the community is still developing or whether it is stable. For the citation network, combining these parameters helps to understand the interdisciplinarity of a conference.
The data set used in our study is the combination of DBLP and CiteSeerX digital libraries. DBLP is a computer science bibliography, which also includes publications in interdisciplinary areas of computer science, including TEL. Additionally the whole database is available for download from the website, easing access to the data. We retrieved the publication lists of conferences from DBLP. However, DBLP does not record citations. Therefore, we used CiteSeerX to fill the citation list of publications in DBLP. DBLP data, as downloaded in March 2011, consists of 881,730 author's names, 1,486,411 publications, 2,868 conference series and 839 journals. The CiteSeerX data set includes 9,121,166 publications, 22,735,140 references and over 6 million author names. We combined DBLP and CiteSeerX using the canopy clustering technique (McCallum, Nigam & Ungar, 2000). Overall, the matching algorithm gave us 864,097 pairs of matched publications. From these data sets we extracted the co-authorship and the citation networks for five main conferences in TEL which are frequently visited by authors of the ET&S journal (see Table 1). The co-authorship networks are based on the DBLP data set. The citation networks are based on the papers and citations in the CiteSeerX data set. Basic statistics regarding the size of the co-authorship and citation networks of these five conference series, including their associated workshops proceedings, are given in Table 2.
Development pattern of TEL conference communities Co-authorship networks
All five conference co-authorship networks are complex networks. In the last seven years, these five conferences combined have published papers written by a relatively stable number of around 1,350 authors each year. For illustration, Figure 3 displays the current co-author network for each of the five conferences in thumbnail form.
[FIGURE 3 OMITTED]
Figure 4 compares the five conferences with respect to the development of the network parameters introduced in the previous section. Figures 4b-f each plot one of the network parameters on the vertical axis, versus the "age" of the conference series on the horizontal axis. The age in this case refers to an ordered series of conference events without taking the actual time interval between events or the point in time of any specific event into account. Age 1 therefore refers the first event in each series, e.g., ECTEL 2006 and ICWL 2002. Age 2 refers to the second event (e.g., ECTEL 2007 and ICWL 2003), and so forth. Since some conference series have a longer history than others, the data series plotted in Figure 4 have a different number of data points. Of course, to compensate for this imbalance, the number of plotted events could be adjusted to the youngest conference series. However, this would impede the comparative analysis of the development patterns of communities of different conference series: to allow for a fair comparison between two conference series from the perspective of a full life-cycle model (Figure 1), we need to compare the community development starting with the inaugural event for each conference series, at age = 1.
The density--i.e., the ratio between the number of edges and the number of nodes--has increased over time with a coefficient larger than 1 and less than 2 for all five conference series (Figure 4a), whereby the coefficient is largest for ITS (1.38) and ECTEL (1.24), and smallest for ICWL (1.05). This means that ITS and ECTEL successfully manage to match the growing set of authors with a growing web of co-authorship connections. ICWL does not perform well in this regard, which is also evident in the plethora of un- or weakly connected small author groups and the absence of a giant component in the network (see Figure 3) even after nine conference events with 425 published papers.
The clustering coefficient (Figure 4b) of all co-authorship networks is quite high (roughly between. 87 and. 92) although it has dropped over the years, but Figure 4d shows that AIED and ECTEL have quickly growing largest connected components (i.e., the core author group described for the development model above) indicating a faster scientific community building process than for ICALT and ICWL. ITS has the largest core author group of all five conferences, but it needed longer to develop, since the size remained at an almost constant low value for the first three ITS events. This might be due to the fact that the first three ITS events indexed on DBLP were held between 1992 and 1998, a time span in which annual conferences would have had seven events. While the plots in Figure 4 do not align the actual points in time of different conference series, it seems safe to assume that the time interval between two consecutive conference events, and the overall temporal continuity of conference events, do have an impact on the development of the conference community. More research is needed to clarify this issue.
For maximum betweenness, ITS also has the highest value at slightly under. 08 (Figure 4c), which means that the most central author in the ITS network is on almost 8% of all shortest paths through the co-authorship network. This value indicates that there are many active key members--i.e., those authors that connect different author communities through co-authoring of papers--contributing to the conference and community development. ICALT and ICWL do not exhibit such a clear pattern, while AIED and ECTEL are developing very fast in this regard. Fast development of the community typically indicates that the conference has a tighter focus and/or the authors publishing at those conferences already had strong ties between each other before the conference series started. ECTEL, for example, is a European conference, so the community is by definition smaller than that of ICALT or ICWL, which address TEL communities worldwide.
All diameters of the co-authorship networks are still growing (Figure 4e), indicating that the development of the community is not yet finished. Since the diameter represents the length of the longest shortest path through the network, a peak in diameter growth would indicate a lack of assimilating new author groups into the core conference community. Also, the average path length is still growing for all five conference series (Figure 4f), indicating again that their networks are still growing.
[FIGURE 4 OMITTED]
To summarize, ICALT and ECTEL have well-connected authors. The remarkable achievement of ECTEL, which is only half as old as ICALT, is likely based in its origins: the conference was started as an initiative out of the EU project PROLEARN in 2006, and to this day remains a strongly EU TEL project-focused presentation outlet and meeting venue. ICWL, on the other hand, is as old as ICALT, and still seems to struggle with managing the transition from the emergence stage to more mature stages in the development pattern shown in Figure 1. The other two conferences--AIED and ITS--exhibit very mature author communities, which is probably due to the fact that these two conferences attract a strong core of artificial intelligence (AI) researchers. In that sense, they are actually difficult to compare with the other TEL conferences, since their core topic is AI rather than TEL.
The citation networks of the five TEL conferences are complex ones with the ratio between the number of edges and the number of nodes still growing (greater than 1 and less than 2 in Figure 5a). The clustering coefficients of all conferences are similar, with ICWL exhibiting a higher coefficient than the other four conferences (Figure 5b). However, Figure 5d shows that the literature of ICWL and ICALT is much less connected than that of ITS, AIED and ECTEL, which indicates that the two former ones have a broader, more interdisciplinary scope than the three latter ones. This is supported by the development of the maximum betweenness values in Figure 5c, which indicate the existence of more common core references in these scientific communities. The diameters of ECTEL and AIED have begun to shrink very early, indicating that the body of literature of these communities is quite stable and the themes of the communities are settled. The development of the average path length also supports this finding.
[FIGURE 5 OMITTED]
Benchmarking TEL conference communities
We compare the development pattern of TEL conferences to the development pattern of four established conferences in database research, including VLDB (Very Large Data Bases Conference), SIGMOD (ACM Conference on Management of Data), PODS (Symposium on Principles of Database Systems) and ICDE (International Conference on Data Engineering). Because of the long history, outstanding reputation and success of the database research community, these conferences can serve as benchmarks or good practice for other conference communities. The network parameters of these conferences over time are given in Figure 6. All four conferences exhibit the same development pattern: they developed steadily from the bonding stage to a focused topology over a timespan of roughly 20 years. After that time, they achieved a stable network, as we see in the fairly stable values of the clustering coefficient, maximum betweenness, diameter, and average shortest path length.
Compared to the development pattern of TEL conferences in Figure 4, we can see that TEL conferences exhibit a pattern typical of "young" communities. Some TEL conferences develop faster, e.g., ITS, AIED and ECTEL, where betweenness and the largest connected component increase very fast, while the clustering coefficient drops and tends to become stable very early. The other two conferences, ICALT and ICWL, have developed more slowly, but they still follow the same pattern. A closer look at the values of the network parameters shows that ICWL (and, to a certain extent, ICALT) clearly faces a challenge: the ICWL community is highly clustered into many unconnected components, thus a giant component (a group of core authors) is missing. In this community, we can see the absence of gatekeepers who connect different groups, as indicated by the low maximum betweenness value.
[FIGURE 6 OMITTED]
Given the comparison of network structures of TEL and database conference communities, and in particular the case of ICWL, we attempt to understand the strategy by which conferences develop their community. In particular, we are interested in the reason behind the emergence--or the absence--of the giant component. One can imagine two reasons for the absence of such a component:
(1) Authors are leaving the conference: if authors publish in the conference once and never come back, they will leave behind "dead" nodes in the co-authorship network, in that they are not active anymore. There will be no connections from these nodes to other nodes in the future. Therefore, new nodes have no chance of connecting to existing nodes.
(2) Authors do return to the conference, but they continue collaborating within their own group. This behavior strengthens the connections within groups, but makes no new connections that cross the sub-communities. The whole community is therefore a set of unconnected groups, which contradicts the very nature of a scientific conference.
The giant component is formed when authors choose to stay with the community and collaborate extensively with other authors. The giant component also becomes bigger when new authors are connected to authors who are already in that component. In both cases, recurring authors play an important role in the development of the community: they ensure the connectivity of the community and their interaction makes the community more cohesive.
To measure the return of authors and their contribution to the community, we calculated the rate of recurring authors and their publications over the years. A paper is published by recurring authors if at least one of the paper's authors has published in the conference before. A high rate of recurring authors, together with a low rate of papers by recurring authors, indicates that recurring authors mainly collaborate with each other (one paper has more recurring authors). On the other hand, a high rate of recurring authors, together with a high rate of papers by recurring authors, indicates that recurring authors collaborate mainly with new authors, which contributes to community development.
[FIGURE 7 OMITTED]
We compared these two measures for the TEL and database research communities (see the plots in Figure 7). As observed, the basic trend during the early stage of the development process is to retain authors: the fraction of recurring authors in all conferences increased in the first years. The frequency of papers by recurring authors also increased. However, keeping this strategy would lead to a closed community and there would be no new ideas coming in from new authors. Therefore, at a certain point in time, conferences retain a healthy fraction of recurring authors. That is one principal strategy for cultivating the scientific community of practice proposed by Kienle and Wessner (2005; 2006). For database communities, we observed this trend in the first 15 years after which the two measures became stable. For TEL conferences, the fraction of recurring authors and their publications in ICWL ceased to increase during the last five years. Year by year, only a small fraction of authors continued to publish in ICWL (less than 25%), while in other TEL conferences, more than 35% authors continued to publish, and this value is still growing. In both TEL and database areas, some conferences quickly managed to retain their authors, e.g., ECTEL and AIED for the TEL community, and PODS for the database community. ICALT currently has the highest fraction of recurring authors, but the fraction of papers by recurring authors is less than at ITS, and this rate has developed comparatively slower during the first couple of events.
Depending on the nature of the conference, there are several reasonable explanations for the above observations. Extremely focused conferences such as AIED can quickly manage to retain a good fraction of their authors since there are not so many prominent options when it comes to publishing in this highly focused field. In more interdisciplinary conferences like ICWL, authors return to the conference at a lower rate. Other practical factors will also have an impact, e.g., the location of the conference venue, the programme committee members, and similar factors. For example, ECTEL until now has been held exclusively in Europe, while PODS was held exclusively in North America in the first 22 years. On the other hand, ICALT has moved across the globe from the beginning, and still manages to retain its authors at a high rate. This suggests that a combination of multiple factors determines how well conference communities manage to keep their members returning.
Discussion and conclusion
In this paper we have explored the structural development of the Technology Enhanced Learning (TEL) scientific community by analyzing the development pattern of co-authorship networks of five major international conference series in TEL which have particular relevance for authors who have also published papers in the ET&S journal: ICALT, ICWL, ECTEL, ITS and AIED. Co-authorship of a research paper is the most explicit demonstration of ongoing and completed collaborative research in a scientific community of practice, and therefore a valuable object of study in this regard. While we used the development of co-authorship networks (in combination with the paper citation networks) as a main factor for explaining the development of conference communities, we acknowledge that there are many additional factors for conference community development such as the reputation of the speakers and audience attending the conference, the quality of papers at previous conferences, the attractiveness of the keynote speakers and co-located events, the location of the conference venue, the organization skills and community-connectedness of the conference chairs, and many more.
We have calculated and compared social network parameters of the conferences series' co-authorship networks by applying a time series analysis to reveal patterns and differences in the community development of these five conference series over time. Overall, all five conference series have developed constantly, though at a different pace. Comparing this pattern with that of established conferences in other sub-disciplines of computer science such as databases, we found that TEL conferences exhibit a development pattern that is typical of young and emerging conference communities. Nevertheless, we see that conferences in TEL are building their community in a way that shapes a clear core. In this sense, maintaining and promoting key members who play the role of gatekeepers and connectors is very important.
We have presented a development model for conference communities, including a sequence of phases that may eventually lead to different co-authorship network topologies. We have shown that highly specialized and focused conferences in computer science tend to develop a focused topology with a very large connected set of authors--i.e., a giant component which, for some conferences, may consist of two-thirds or more of all authors in the conference series' history (e.g., VLDB or ACM SIGMOD). One key feature of TEL, as opposed to highly focused conferences, is its interdisciplinary nature. In the analysis of conference community development patterns, it became evident that interdisciplinarity comes with pros and cons. On the one hand, it attracts researchers from different subject areas to a conference. On the other hand, it slows down the process of building a core group of authors, as we saw, for example, in the relatively slow development pattern of ICWL compared to faster developing, more focused conferences like AIED and ITS, which have a clear artificial intelligence focus, or ECTEL, which has its geographic and thematic focus in European TEL. By far the most important conference series for ET&S authors is ICALT: in the events since 2005, ICALT has attracted an average of over one hundred papers annually by authors who have also published in the ET&S journal.
An analysis of continuity of authorship, i.e., authors who publish in more than one event in the conference series, shows that in the early stages, conferences build their community by retaining authors from previous events. The returning authors and their contributions (papers) are a key driver for the development of a large, well-connected core author group that is characteristic of mature conference communities. However, pushing too hard in this direction would close the community, and there would be no more new ideas coming in the form of work by new authors. Conference organizers thus need to strike a balance between measures to retain authors on the one hand, and measures to attract new authors on the other. One potential way of achieving this is to move the conference to a different place (e.g., another continent) every once in a while, since this will attract new, local researchers plus a share of the conference's veterans. The key success factor appears to be the integration of the established conference community with the newly attracted authors. If we look at ICWL, for instance, there is no clear sign of a large coauthor group in the network although the conference has recently celebrated its tenth anniversary. One explanation could be that the conference locations started alternating between Europe and Asia after six venues exclusively located in the Asia-Pacific region, thus posing the challenge of connecting two geographically separated communities after the conference moved to Europe for the first time. This may take more time than expected. On the other hand, ICALT has moved across the globe from the very beginning, and still it managed, during the same time span, to attract more tightly connected authors. This indicates that new authors have been successfully assimilated by the ICALT community.
Understanding the community helps the members to define strategies to support its development. One main goal is to move the community toward a focused topology of connections which will offer a fertile scientific environment for research collaboration. Drawing from the analyses in this paper, some recommendations to different conference community stakeholders for contributing toward this goal are summarized in the following paragraphs.
For conference organizers, besides managing organizational issues such as maintaining stable and reputable committees, or moving the conference to opportune locations to attract and involve local researchers, their efforts in retaining the key authors of the conference are very important for the development of the community. Key authors can be rewarded and attracted in several ways, e.g., through offering them roles in the organizing committees or through opportunities for plenary addresses (e.g., keynotes) or similar occasions where they can spark future cooperation by sharing their work and vision.
The key members of the community, i.e., those with a high centrality in the co-authorship network, can contribute to community development not only by publishing papers, but also by cultivating communication between current and prospective community members. With their knowledge of the conference topics and community, they should be active in finding, suggesting and setting up new collaborations with members in different sub-communities, particularly from the conference network's periphery, which will make the whole community more integrated and cohesive. Normally, key members are positioned at the interface between sub-communities, so they are aware of the information and ideas emerging from different sources. Subsequently they can synthesize these sources of information to generate ideas and gather together authors from different sub-groups to work on these ideas. Key members also play an important role in engaging new authors and connecting them to the core of the community. This can also lead to the introduction of new ideas and research topics to the conference.
Finally, for all other conference community members, continuing work in their established co-author sub-community will strengthen existing collaboration ties, but may impede the development of the whole conference community. Engaging in collaborations that span different sub-communities helps to generate collaboration ties within the conference community, and can also strengthen the reputation of authors who are sparking these collaborations. In terms of community topology, this also contributes to making the community more cohesive and focused. Additionally, connecting different sub-communities is an important indicator for the future reputation of these authors and their status as "gatekeepers" based on their centrality in the network.
In future work we are planning to augment the findings drawn from structural analysis of the conference communities with semantic analysis of papers published at conferences by different authors and author groups. With this structural-semantic analysis we expect to be able to recommend authors for collaboration and papers for reading to community members. It will also provide more insight into the development of the thematic focus of conferences and the roles of key authors.
This work has been supported by the BIT Research School of RWTH Aachen University and University of Bonn, and by the EU FP7 projects TEL-Map (http://telmap.org) and ROLE (http://role-project.eu).
Boyack, K.W., Borner, K. & Klavans, R. (2007). Mapping the structure and evolution of chemistry research. In: D. Torres-Salinas & H. Moed (Eds.), Proceedings of the 11th International Conference on Scientometrics and Infometrics (pp. 112-123).
Ding, Y., Chowdhury, G., & Foo, S. (2000). Journal as markers of intellectual space: journal co-citation analysis of information retrieval area. Scientometrics, 47(1), 55-73.
Hoadley, M. C. (2005). The shape of the elephant: scope and membership of the CSCL community. In T.-W. Chan (Ed.), Proceedings of the 2005 Conference on Computer Support for Collaborative Learning, CSCL '05 (pp. 205-210). International Society of the Learning Sciences.
Kienle, A. & Wessner, M. (2005). Our way to Taipei: an analysis of the first ten years of the CSCL community. In T.-W. Chan (Ed.), Proceedings of the 2005 Conference on Computer Support for Collaborative Learning, CSCL '05 (pp. 262-271). International Society of the Learning Sciences.
Kienle, A. & Wessner, M. (2006). Analysing and cultivating scientific communities of practice. International Journal of Web Based Communities, 2(4), 377-393.
Leskovec, J., Kleinberg, J. & Faloutsos, C. (2007). Graph evolution: Densification and shrinking diameters. ACM Transactions on Knowledge Discovery from Data, 1(1), Article 2.
McCain, K.W. (1998). Neural networks research in context: a longitudinal journal cocitation analysis of an emerging interdisciplinary field. Scientometrics, 41(3), 389-410.
McCallum, A., Nigam, K. & Ungar L.H. (2000). Efficient clustering of high-dimensional data sets with application to reference matching. In Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 169-178). New York: ACM.
Morris, T.A. & McCain, K.W. (1998). The structure of medical informatics journal literature. Journal of the American Medical Informatics Association, 5(5), 448-566.
Ochoa, X., Mendez, G. & Duval, E. (2009). Who We Are: Analysis of 10 Years of the ED-MEDIA Conference. In G. Siemens & C. Fulford (Eds.), Proceedings of World Conference on Educational Multimedia, Hypermedia and Telecommunications 2009 (pp. 189-200). Chesapeake, VA: AACE.
Pham, M. & Klamma, R. (2010). The structure of the computer science knowledge network. In N. Memon & R. Alhajj (Eds.), International Conference on Advances in Social Networks Analysis and Mining, ASONAM 2010 (pp. 17-24). Los Alamitos, CA: IEEE Computer Society.
Pham, M., Klamma, R. & Jarke, M. (2011). Development of computer science disciplines: a social network analysis approach. Social Network Analysis and Mining, 1(4), 321-340.
Reinhardt, W., Meier, C., Drachsler, H. & Sloep, P. (2011). Analyzing 5 Years of EC-TEL Proceedings. In C. Kloos, D. Gilet, R. Crespo Garcia, F. Wild & M. Wolpers (Eds.), Towards Ubiquitous Learning-6th European Conference of Technology Enhanced Learning, EC-TEL 2011 (pp. 531-536). Berlin: Springer-Verlag.
Tsay, M.Y., Xu, H. & Wu, W.C. (2003). Journal co-citation analysis of semiconductor literature. Scientometrics, 57(1), 7-25.
Wasserman, S. & Faust, K. (1995). Social Network Analysis: Methods and Applications. Cambridge: Cambridge University Press.
Wild F., Haley, D. & Buelow, K. (2010). CONSPECT: monitoring conceptual development. In X. Luo, M. Spaniol, L. Wang, Q. Li, W. Nejdl, W. Zhang (Eds.), Advances in Web-Based Learning-ICWL 2010-9th International Conference (pp. 299-308). Berlin: Springer-Verlag.
Manh Cuong Pham, Michael Derntl and Ralf Klamma
RWTH Aachen University, Advanced Community Information Systems (ACIS), Aachen, Germany // email@example.com // firstname.lastname@example.org // email@example.com
Table 1. Conferences relevant to the ET&S journal Conference Series Acronym Series Events IEEE International Conference ICALT Annually 2001-2010 on Advanced Learning Technologies (except 2002) Artificial Intelligence in Education AIED Bi-annually 2005-2009 European Conference on Technology ECTEL Annually 2006-2010 Enhanced Learning International Conference on ITS Bi-annually 1992-2010 Intelligent Tutoring Systems (except 1994) International Conference on ICWL Annually 2002-2010 Web-Based Learning Conference Series Relevancy * IEEE International Conference 103 on Advanced Learning Technologies Artificial Intelligence in Education 31 European Conference on Technology 28 Enhanced Learning International Conference on 14 Intelligent Tutoring Systems International Conference on 10 Web-Based Learning * ... Average number of papers per event in the conference series which were written by authors who have also published in ET&S. Focusing on recent papers, we only considered conference events since 2005 for calculating this metric. Table 2. Co-authorship and citation network statistics for the selected conferences Conference # Papers Co-Authorship Network Citation Network Series # Nodes # Edges # Nodes # Edges ICALT 2,299 4,021 7,222 905 618 AIED 535 961 2,220 1,055 2,021 ECTEL 450 940 1,988 812 596 ITS 976 1,627 4,234 1,977 4,239 ICWL 425 945 1,518 325 273
|Printer friendly Cite/link Email Feedback|
|Author:||Pham, Manh Cuong; Derntl, Michael; Klamma, Ralf|
|Publication:||Educational Technology & Society|
|Date:||Jul 1, 2012|
|Previous Article:||Student engagement in blended learning environments with lecture-based and problem-based instructional approaches.|
|Next Article:||Flexible Pedagogy, Flexible Practice: Notes from the Trenches of Distance Education.|