Advertisement

Springer Nature is making SARS-CoV-2 and COVID-19 research free. View research | View latest news | Sign up for updates

Investigating collaboration in ubiquitous computing research

Abstract

With the rapid development of mobile internet and smart devices, ubiquitous computing is experiencing a huge change. As direct progress reflection of the field, collaboration among scholars has been increasingly important and facilitating a few emerging applications, e.g., collaborator recommendation. In this paper, to further figure out how the contributors collaborate with each other, we explore two representative conferences in ubiquitous computing research field, UbiComp and Pervasive (merged with UbiComp in 2013). Papers collected from the two conferences are data set to examine collaboration from three perspectives: (1) Collaboration overview: we construct collaboration network and conduct analysis from aspects of author, institution, and nation respectively, and shed light on progress of the field. (2) Collaboration Evolution: by segmenting the whole time period of collaboration into two distinct stages, we identify different kinds of collaboration patterns in terms of collaborator and research topic. (3) Collaboration Quality: we propose a new method to quantify the actual quality of each kind of collaboration patterns. Furthermore, we propose some suggestions to development of ubiquitous computing according to our findings.

Introduction

Over the past few decades, ubiquitous computing has grown into an active field covering mobile sensing, crowd sensing, context-aware computing, etc., which is attracting more and more scholars and has integrated relevant institutions into a stable community. As direct interaction within the field, research collaboration can be seen as a process of knowledge-sharing as well as facilitating the achievement of common goals. Previous studies have provided us with a few insights of research collaboration, and meanwhile facilitated the emergence of a set of applications, e.g., collaborator recommendation, collaboration prediction, and expert finding Kronegger et al. (2012). In this article, to figure out collaboration of ubiquitous computing, we use published papers collected from UbiComp and Pervasive (merged with UbiComp in 2013) covering from 1999 to 2018 as data set and adopt co-authorship network as indicator of research collaboration. We then examine collaboration from three perspectives, (1) Collaboration overview, (2) Collaboration evolution, and (3) Collaboration quality.

We first look in detail at the collaboration network and conduct analysis from structural aspect. A collaboration network is established in which the authors correspond to vertices, and collaborations between different authors correspond to edges with weights representing the number of coauthored papers. Specifically, by analyzing graph features, e.g., giant component, degree, and cluster coefficient, an overview of collaborations in ubiquitous computing is presented. Afterwards, to examine collaboration evolution, we propose a principal component analysis (PCA)-based approach to deal with the time series segmentation. Compared with existing studies either choosing segmentation arbitrarily or according to the change of topics Zhang et al. (2010), this method segments time series in terms of author importance and suggests 2008 as a break point of the field. Then, based on two periods (1999–2008 and 2009–2018) segmentation, we investigate different collaboration patterns in terms of collaborators and topics and further evaluate their impacts on collaboration. Moreover, we define the collaboration quality, which is measured by considering both the number and the citation amount of coauthored papers, and we further evaluate it in different kinds of patterns and make our suggestions. As a result, we reveal several collaboration patterns and quantitatively evaluate their impact on research collaboration, which could shed light on development of the field. Then we make a discussion about the interesting findings and results of experiments. Furthermore, we provide hints for building better collaboration relationships to maintain the development of academic field sustainable and healthy. For example, holding the conference in different countries will benefit enlarge influence of UbiComp and attract more scholars construct creative work in the field. Finally, we make a conclusion that in order to promote the quality of individuals’ research work, scholars are supposed to first extend collaboration with others, and second conduct in-depth research when finding a new exploring topic. To sum up, the main contribution of this work is fourfold:

  • We make a collaboration overview in different aspects by constructing and analyzing collaboration networks across authors.

  • Based on the proposed PCA-based time series segmentation approach, we identify four types of collaboration patterns including cohesive pattern, volatile pattern, topic unchanged pattern and topic changed pattern, and systematically analyzed the collaboration of UbiComp in terms of collaboration evolution.

  • We propose a new collaboration quality evaluation method for the scientific literature combining citations and the number of published papers. And we quantified the four kinds of collaboration patterns based on this proposed method.

  • We provide hints for scholars to promote their research works according to the findings and results of experiments, which could help researchers to build better collaboration relationships and maintain sustainable development of academic field.

The remaining of this paper is organized as follows. We first review related works in Sect. 2. And in Sect. 3, we build collaboration networks and present the overview of UbiComp’s collaboration in two aspects during the past 20 years. In Sect. 4, we propose the PCA-based time series segmentation scheme and make evolution analysis putting forward four types of collaboration patterns. Collaboration quality is defined in Sect. 5 and we evaluate impacts of different collaboration patterns using such metric. Discussions about how to build better collaboration relationships to maintain the development of academic field sustainable and healthy are given in Sect. 6. And finally, we conclude the paper in Sect. 7.

Related work

In this section, we will review the related works from two categories that best line up with our research: collaboration network analysis and application research based on co-authorship network.

Collaboration network analysis

The first category describes collaboration network from the perspective of graph theory, which is the very foundation of our work. Social network is a relatively stable system of relations, which is constituted by a group of interrelated and interactional entities Klemmer et al. (2002). As a subcategory of social network, the scientific collaboration network is of importance and has been widely discussed in the literature, Fung and Wong (2017) examined publication output and co-publication network structures to investigate the comparative advantage and composition of the research networks in the various economies. As the direct case resulting in coauthored publications, co-authorship has been widely adopted as the best indicator of research collaborations. Newman (2001) constructed a co-authorship network using scientific papers in physics, biomedical research, and computer science and found a number of network differences between experimental and theoretical disciplines. In addition, by presenting multi-dimensional statistical properties of co-authorship networks, Newman (2004) also answered a broad variety of questions about collaboration patterns and provided an overview of scientific collaboration. Zhang et al. (2010) collected 25 year’s published papers in IEEE Intelligent System and identified the key collaboration facilitator using three different measures namely degree, betweenness and PageRank. Instead of partitioning time series randomly, this paper divided time series into two periods by applying two basic time series segmentation algorithms which is capable of revealing the vital time point corresponding to the development of the field and gave us a hint to further study this issue. Barabâsi et al. (2002) analyzed the evolution of scientific collaborations in mathematics and neuro-science and found that the network evolution is governed by preferential attachment. Coccia and Wang (2016) found a convergence of long-run collaboration patterns between the applied and basic sciences by analyzed scientific collaboration network.

All of these studies worked as the foundation for our paper and provided us with insights of scientific collaboration. However, most of them fixed on macro effect of collaboration, few has ever considered specific patterns, which is one of the motivations for this paper.

Application research based on co-authorship network

With the worldwide academic communication getting prosperous, the research on promising applications and systems is attracting increasing attentions. Zeng et al. (2010) presented a coauthor network topic model constructed based on Markov random fields and showed that the higher-order relations among coauthors can improve the topic and expertise modeling performance. Different from most of existing studies treating co-authorship as a homogeneous network, Sun et al. (2011) implemented co-authorship prediction by constructing a heterogeneous network incorporating multiple types of objects, e.g., venues, topics, papers and multiple types of links among objects. In addition, Chen et al. (2011) introduced CollabSeer, an open system to recommend potential research collaborators for scholars which is measured by relation strength, vertex similarity and lexical similarity. CollabSeer considers both the structure of a co-authorship network and an author’s interests for collaborator recommendation. Lopes et al. (2010) considered both the semantic issues involving the relationship between the scholars in research area and the structural issues by analyzing the existent relationships among researchers to make collaboration recommendations. In another research, Huynh et al. (2013) proposed the trend factor which takes temporal change into account to improve the measurement of vertex similarity. Specially, given the existence of isolated researchers (i.e., researchers who have no links with others in co-authorship network), Huynh et al. (2014) utilized additional information, e.g., the strength of the relationship between organizations, as new features to make recommendation. In general, no matter what purpose of a research is, the vertex similarity referred to the relationship between nodes, has been utilized as the fundamental indicator to characterize co-authorship network. In addition, the practical utility of co-authorship proved by the preceding studies prompts us to further investigate the way the UbiComp collaborates, which will contribute to the recognition of conference and meanwhile unveil the growing trend in the future.

Collaboration overview

In this section, we present collaboration overview from three perspectives, namely author, institution and nation. Our dataset is collected from DBLP bibliography for UbiCompFootnote 1 and PervasiveFootnote 2 including 1521 published papers and 3774 distinct authors from 1999 to 2018.

Author-collaboration

To investigate author-collaboration, we extract authors of each paper as entities and manually handle the disambiguation of names such as the same name and abbreviation to assure identities. Then we construct a collaboration network in which vertexes represent authors. If two authors have collaborated with each other, there is an edge between the correspondent vertexes, and the weight of edges represents the number of papers the two authors have published together. Figure 1 illustrates the profile of constructed network, where the bigger nodes represent more productive authors. For a clear presentation, we just show a part of author’s name in the figure. Accordingly, the backbone of the field is presented, and the profile provides us with vision of collaboration as well as a new view to inspect prominent researchers. The collaboration network is comprised of two types of clusters. One is a giant component, i.e., the largest connected component, which dominates more than half of the network. The other one consists of plenty of relative isolated groups, part of which may only have published one paper. Thus, the giant component not only represents the foundation but also reflects the development of the field properly. However, the collaboration can still be diverse even the giant component seems to be unchanged. We conduct a survey on ratio of new authors which have never appeared in the conference before for each year. The result is shown in Fig. 2, and it is obviously the ratio of new authors is increasing yearly. It means collaborations are diverse and the field is attracting more and more scholars to participate in collaborating. Specifically, the average degree represents the number of distinct collaborators an author has collaborated with, while the average weighted degree represents the average number of collaborations. The distribution of average degree is demonstrated in Fig. 3, where plenty of isolated authors constitute the rising part at the beginning and the declining trend since five fits the power-low distribution. It means that collaboration is usually about five people more or less.

Fig. 1
figure1

The collaboration network of authors

Fig. 2
figure2

The ratio of new authors each year

Fig. 3
figure3

The distribution of degree

Moreover, we also make a comparison with some other conferences of computer science. We similarly draw the collaboration network of three other fields, namely KDDFootnote 3 (papers of 1995–2018), CIKMFootnote 4 (papers of 1992–2018) and WWWFootnote 5 (papers of 2001–2018) and compare them with ubiquitous computing using several general networking measures. The statistic results are shown in Table 1. N is the number of authors, PG is the percentage of authors in giant component, AD is the average degree, NCC is the number of connected components, CC is the clustering coefficient and it is calculated as:

$$\begin{aligned} CC = \frac{3 \times \text {Number}\_\text {of}\_\text {triangles}}{\text {Number}\_\text {of} \_\text {connected}\_\text {triples}}. \end{aligned}$$
(1)

where a triangle represents trios of vertexes in which each vertex is connected to each others, while connected triples represents trios of vertexes for which at least one vertex is connected to the others Bian et al. (2014). Specifically, the average degree represents the number of distinct collaborators an author has collaborated with and clustering coefficient reflects the gathering degree. While the PG and the AD of ubiquitous computing are all in the middle, the lower NCC and greater CC means the more cohesive collaboration.

Table 1 Statistics of the collaboration network of different research fields

Affiliation-collaboration

Within the 1521 published papers, 583 institutions are extracted and classified into three categories, namely university (64.3%), corporation (12.9%) and institute (22.7%). We identify different institutions by extracting keyword of their names and manually deal with abbreviations such as MIT and KAIST. Similarly as the collaboration network presented in Fig. 1, we further construct an institution-collaboration network in Fig. 4, where the red dot represents university, yellow dot represents corporation and blue dot represents institute, the size of node indicates number of collaborations. It is worth noting that we extract the institution given by the authors in the paper they published and choose the first one if the author belonged to more than one institution at that time. Therefore, we are able to avoid the institution-switched problem and make sure that an author will not be matched to different institutions at the same time. In Fig. 4, most nodes are connected and collaborations of three categories are interlaced. It is significant that many ideas will be proposed and they have a chance to be applied to improving our daily life and society during the collaborations of institutions especially between university and corporation. Meanwhile, practical demands are able to push the development of the field.

Fig. 4
figure4

Institution-collaboration network

Similarly, based on author’s nationality, we construct nation-collaboration network in Fig. 5. We mark countries with different colors (light blue as North America, red as Europe, yellow as Asia, blue as Oceania, black as Africa, pink as South America). We can find from the figure that USA is the most productive country in ubiquitous computing, and Europe takes prominent part of network. Among top 20 of countries, Europe takes proportion of 65% (13 countries), UK and Germany is in the second and third place respectively. As for Asia, Japan (4th), China (6th) and Korea (8th) are top three counties. The most interesting finding is that the top countries all have held UbiComp conference, and the number of papers produced by authors from these countries (especially Asia) is increasing obviously after the country held the conference. Therefore we have reasons to believe that changing holding area of conference is benefit to expand influence of UbiComp and is also effective to promote the collaboration between different countries.

Fig. 5
figure5

Nation-collaboration network

Collaboration evolution

In this section, we will investigate the evolution of collaboration from two aspects: time and collaboration pattern. We first introduce a PCA-based approach to deal with time series segmentation, and then we analyze the evolution process and characterize different collaboration patterns.

Time series segmentation

Twenty year is not a short history, and scholars of the field have been changing along with the change of research interests. To examine the evolution of collaboration, we need to segment the time series. On one hand, the evolution of collaboration is a long process along with slight changes every year. A proper segmentation could point out the most representative time that reflects the change of collaborations. On the other hand, a reasonable segmentation could increase efficiency rather than random analysis. In this paper, instead of arbitrarily separating time series, we propose a PCA-based scheme to find appropriate period as the basis of further analysis.

As mentioned above, the collaboration network is consist of several connected components with only one of them governing more than half of the network, which explicitly reflects different structural impact of each subcomponent. As we can see, during the process of development, there exist authors who permanently make contributions to the conference as well as authors just working as peripheral men. Obviously, the former groups are core authors, who should be regarded as the backbone and can better represent the history of the field. In our segmentation method, we leverage the core authors each year and separate the time series according to the change of quantity of core authors.

For each paper, we apply 3,774 authors as features to represent it as a vector. The vector of the paper \({P_n}\) is defined as : \({P_n} = \left\langle {\begin{array}{*{20}{c}}{{a_1}}&{{a_2}}&\cdots&{{a_{3774}}}\end{array}} \right\rangle\). The \({P_n}\) represents the n-th paper, and \({a_1}\) to \({a_{3774}}\) corresponds to 3,774 authors. If the author \({a_i}\) has participated in the paper \({P_n}\), the value of \({a_i}\) is 1 and the value of it is 0 on the contrary. Assume that the number of published papers in j-th year is m, then the paper-author matrix \({Y_j}\) of 3774 dimensions is formally defined as follows:

$$\begin{aligned} {Y_j} = \left( {\begin{array}{*{20}{c}} {{P_1}}\\ {{P_2}}\\ \vdots \\ {{P_m}} \end{array}} \right) \begin{array}{*{20}{c}} {}&{\left( {j = 1,2, \ldots ,20} \right) }. \end{array} \end{aligned}$$
(2)

Accordingly, we build 20 matrixes to denote author-paper relation of UbiComp during 20 years.

Then, given the sparsity of the matrix, which is caused by limited number of publications and researcher turnovers, i.e. there are plenty of zero value, we apply principal component analysis (PCA) to process the matrix Korada et al. (2011), to find the core authors, i.e. the most representative authors. PCA performs dimensionality reduction based on the eigenvector of covariance matrix and in consequence, generates fewer new features (i.e. core authors each year). We empirically set 0.95 as the threshold to filter the core authors, and the result is presented in Fig. 6 where Y-axis corresponds to the number of core authors and X-axis corresponds to 20 years. Then we apply Bottom–Up Algorithm Lopes et al. (2010), which iteratively combines consecutive time segments according to their differences. It is clear that the increasing speed of slope of the curve after 2008 is faster than that before 2008. This method finally segments time series into two periods, i.e. 1999–2008 as Period I and 2009–2018 as Period II.

Moreover, based on the principle of PCA, the core authors extracted here work as nodes connecting more authors as well as publishing more papers. Hereby, we believe that the change of core authors is able to essentially reflect the change of the conference in terms of collaboration and research interest. To further understand the effect of core author changes, we compute the ratio between the number of core authors and the number of total authors. The result is shown in Fig. 7. We can see that, though the absolute number of core authors is increasing, the ratio decreases as time passing by. In other words, in the early period there were few authors participate in UbiComp so almost of them could be regarded as core authors, but with the development of the field, more and more researchers participate in UbiComp and relative fewer authors are connecting all authors together, which indicate the enhancement of collaboration.

Fig. 6
figure6

The change of core authors’ number

Fig. 7
figure7

The change of core authors’ ratio

Evolution analysis

The collaboration overview in Sect. 3 intuitionally provides us with knowledge from a static aspect. However, the collaboration itself is a process with establishment of new relationships and maintenance or dissolution of old ones, which can be directly reflected in the patterns of collaboration. To further understand this process, it is imperative to study the specific ways that collaborations happen.

Collaboration pattern about continuity

Actually, the change of collaboration is the change of co-authorship indeed, which inspires us to inspect collaborators for understanding the collaboration pattern. During the academic career of scholars, different kinds of collaborations would happen in terms of different organizations and areas, i.e., domestic and international, etc. From the perspective of continuity, collaborations could be classified into two categories, multi-time with the same person and limited time with different persons, respectively. The difference between these two types is essentially dependent on the researchers, which reflects personal preferences, the research background, and even personalities. In other words, there are mainly two types of scholars according to these two distinct collaboration patterns, including those who keep changing collaborators and those who retain a stable co-authorship. According to this property, we refer the former or the later type of collaboration as volatile or cohesive pattern of collaboration. We identify the two patterns based on the time series segmentation. Specially, if collaborators of an author in the two periods are extremely different, we consider the collaboration pattern of the author as volatile pattern, while that is cohesive collaboration pattern on the contrary. In order to discover such collaboration patterns, it is essential to collect scholars who have published one paper at least in both of the two time periods. And as a result, we totally extract 175 authors who satisfy this requirement in our experiments.

Given that the collaboration is presented by an author and his direct neighbors in the network, it is natural to calculate the similarity between neighbors of an author across two time periods to qualify the continuity of collaboration. Among many implemented approaches such as Jaccard similarity, cosine similarity and topology overlap similarity, cosine similarity has been proved to be better than Jaccard similarity in sparse data. Considering the sparsity of the collaboration compared with abundant number of scholars, we select classical cosine similarity Luo et al. (2008) to denote the degree of collaboration continuity for each author in two periods. And the formal definition for determining the degree of continuity is presented as Eq. 3:

$$\begin{aligned} C{T_i} = \frac{{\left| {{N_{i,1}} \cap {N_{i,2}}} \right| }}{{\sqrt{\left| {{N_{i,1}}} \right| \left| {{N_{i,2}}} \right| } }}. \end{aligned}$$
(3)

\(C{T_i}\) represents cohesive trait of an author i, \({N_{i,1}}\) is the collaborator set in Period I (1999-2008), and \({N_{i,2}}\) is the collaborator set in Period II (2009-2018). \(\left| {{N_{i,1}}} \right|\) and \(\left| {{N_{i,2}}} \right|\) respectively correspond to the collaborators’ number of the set in two periods. The result is shown in Fig. 8, where the higher similarity indicates higher collaborator continuity. The exponential distribution reveals the existence of two collaboration patterns and the cohesive pattern is not a trivial phenomenon. Moreover, since the collaboration pattern is rather a relative concept, we empirically set similarity of 0.2 as the threshold to classify author’s collaboration patterns into volatile or cohesive respectively.

Fig. 8
figure8

The similarity of collaboration in two periods

Collaboration pattern about topic

Besides the changing of collaborators, when we discuss scientific collaborations, it is natural to think of what role the research topic played and how the topic influences the way people collaborate during the evolution of topics. Abundant existing studies have examined this issue from different aspects (Jensen et al. 2015; Masada and Takasu 2012). Over the past 20 years, topic evolution greatly affected the pace of UbiComp and extended the coverage of the conference from location-aware applications, networks, crowdsensing and data analysis Liu et al. (2014). With the appearing of new topics and fading of old ones, the scholars keep making their way by insisting on digging deeper in the same field or following the trend and exploring new topics, both of which are indispensable to make the conference stepping forward. The former one intuitionally benefits from profound knowledge of a specific area while the latter one takes advantage of feasibility and innovation. In contrast with traditional research areas, the hot topic of UbiComp changes quickly because of rapid development of intelligent devices and networks during the past decades, which to some extent prompts researchers to explore new areas and simultaneously intensifies the difference of two research patterns. Nevertheless, how to evaluate the different impact and how the difference is reflected in collaboration still remains a mystery. Therefore, we attempt to evaluate whether an author has been fixed on the specific topic, which we refer to topic continuity and it is also a relative concept like collaboration continuity.

To the best of our knowledge, although previous studies have paid much attention on topic detection Yap et al. (2006), and topic evaluation Jayabharathy et al. (2011), there is no existing criterion to measure academic topic specificity in terms of researchers. In our work, based on textual information of 1521 papers, we apply topic similarity White et al. (2004) as the indicator to describe the topic evolution of authors. We take into account both title and abstract of each paper as textual sources to model the topic preference of authors in different periods and evaluate topic similarities accordingly.

We first extract titles and abstracts of all the papers, and utilize LDA Liu et al. (2014) to construct topic models. Certainly, we have removed the stop words like conjunctions, preposition, and some nouns (such as paper, work, etc.), which appear constantly in papers but have no ability to reveal the main content of papers. Therefore we can attain more brief text to construct more accurate topic model. Finally, there are 15 topic queues composing the topic model, and the result is shown in Table 2.

Table 2 Statistics of the collaboration network of different fields

Similar to discovering the collaboration continuity pattern, we need to focus on those authors who have published papers in both two time periods. Specifically, we collect authors who exist in both two time periods and have published more than four papers in our experiments. Finally, there are 145 authors being extracted from the dataset. We then separately extract keywords of publications in the two periods based on the topic model referred in Table 2. From the point of mathematics, an author’s achievements can be formally represented as a vector of keywords as formula 4:

$$\begin{aligned} {T_i} = \left\langle {{k_{i,1}},{k_{i,2}},{k_{i,3}}, \ldots {k_{i,n - 1}},{k_{i,n}}} \right\rangle . \end{aligned}$$
(4)

\({k_{i,1}}\) represents how many time the n-th keyword appeared during period \({T_i}\). The distribution of the keyword vector is a direct reflection of an author’s interests. Depending on the time series segmentation, two vectors are built for each author. Afterwards, we apply cosine similarity to compute the topic similarity in order to divide collaboration into topic changed pattern and topic unchanged pattern. The formal definition of topic similarity \(Sim\left( {{T_1},{T_2}} \right)\) is presented as formula 5 which is based on the intuition that two vectors are more similar if they share more common keywords:

$$\begin{aligned} Sim\left( {T{}_1,T{}_2} \right) = \frac{{\sum \nolimits _{i = 1}^n {{T_{1,i}} \times {T_{2,i}}} }}{{\sqrt{\sum \nolimits _{i = 1}^n {{{\left( {{T_{1,i}}} \right) }^2}} } \sqrt{{{\sum \nolimits _{i = 1}^n {\left( {{T_{2,i}}} \right) } }^2}} }} \end{aligned}$$
(5)

\({T_1}\) indicates the topic vector in Period I, \({T_2}\) in period II. \({T_{i,j}}\) represents the appearance time of a specific keyword. Obviously, a higher similarity means that the author has been more concentrated on one field. The result of topic similarity of 145 authors between the two periods is shown in Fig. 9. It reveals that as most of the authors keep finding new topics (left part), there exists groups focusing on similar topics (right part) and meanwhile confirms the existence of two patterns, i.e. topic changed pattern and topic unchanged pattern. Similar to collaborator similarity, topic similarity is also a relative concept and we empirically set 0.2 as threshold to classify authors.

In the next section, we will evaluate effect of the four collaboration patterns by defining and comparing quality of collaboration.

Fig. 9
figure9

The similarity of collaboration in two periods

Collaboration quality

In this section, we quantitatively measure the quality of different collaboration patterns utilizing a quantization parameter we named collaboration quality. Although there have been studies investigating the effect of papers involving citation and other factors, most of them focused on single or limited instances and treated it as a static issue Shi et al. (2010). In our work, from the perspective of collaboration, we take temporal evolution and cumulative effect into account to quantify the actual effect of papers in different periods and further evaluate the quality of collaborations.

As we know, papers are productions of collaborations, so we are supposed to utilize characteristic of papers to represent quality of collaboration. To formally define the effect of collaboration patterns, we mainly consider two factors: (1) the number of published papers in the collaboration pattern and (2) the citation amount of published papers, which is normally used to evaluate paper effect (Lu and Dietmar 2010; Ch’Ao et al. 2016). The former can provide intuitional information about the ability of authors while the latter accurately evaluate the effect of papers, both of which have been proved efficient and used in many studies. However, there is an undeniable fact that the earlier published papers are always more likely to be cited than subsequent ones. Meanwhile, it is undoubted that while most recently published papers are barely cited, it does not deduce the low quality of those papers. To avoid bias of cumulative effect, we consider a gradient model to equalize the actual effect of different citations.

We collect the citation amount of each paper from Google Scholar, and have discovered that the change of average citation amount can be properly fitted to a linear trend. Therefore we use slope of fitting curve as gradient parameter, and compute quality of collaboration by combining quotient of coauthored papers and equalized citations. Figure 10 shows the distribution of average citations each year, e.g., dividing the sum of citation per article in the year by the total number of papers in that year. The curve we have drawn fits a trend of linear variation properly with the slope as minus 10.6. By applying this damping factor onto subsequent papers, the cumulative effect is eliminated and we then use the relative value of citations to denote the actual effect of papers. Accordingly, the quality of collaboration for author is formally defined as Eq. 6:

$$\begin{aligned} Q{C_i} = \ln \left( {\frac{{{C_{i,2}} - \alpha {T_i}}}{{{C_{i,1}}}} + 1} \right) \cdot \frac{{{P_{i,2}}}}{{{P_{i,1}}}}. \end{aligned}$$
(6)

\(\alpha\) represents the damping factor of minus 10.6, \({C_{i,1}}\) and \({C_{i,2}}\) represent the number of citations in period I and period II respectively. \({P_{i,1}}\) and \({P_{i,1}}\) denote the number of published papers. \({T_i}\) describes the average time interval which is computed as the difference between middle year of Period I and that of Period II. Logarithmic function is applied to generalize order of magnitudes.

Fig. 10
figure10

The average citation in each year

In Sect. 4, we have defined four kinds of collaboration patterns in two aspects. For collaboration continuity, we divide 175 authors into two patterns, e.g., volatile pattern and cohesive pattern. Similarly, for research topic, we divide 145 authors into two patterns, e.g., topic changed pattern and topic unchanged pattern. Afterwards, we calculate collaboration quality of these four collaboration patterns utilizing Eq. 6 we proposed above, and compare the results separately.

As for collaboration continuity, the distribution result of collaboration quality is demonstrated in Fig. 11, where the CDF of cohesive pattern and volatile pattern present a significant difference. Under the cumulative probability of 60%, the quality of volatile pattern is lower than 2.5 while cohesive pattern reaches around 4. In addition, the curve of the cohesive pattern lasts to 14 compared with 7 of the volatile pattern, which suggests the cohesive pattern is more inclined to result in superior achievements. In addition, we also compare the giant component with isolated cluster, and the curve illustrates that the quality of collaboration among the isolated cluster is the lowest, and the curve lasts only to 3. Based on all of these evidences, we conclude that authors of giant component are more productive and cohesive pattern generally outperform that of the volatile ones.

Fig. 11
figure11

The quality evaluation of three patterns about collaboration continuity

As for collaboration about research topic, the evaluation result is shown in Fig. 12, where three curves overlap at the beginning and separate round cumulative probability of 2.2. The quality of collaboration under topic unchanged lasts to more than 9 while the quality of collaboration under topic changed is lower than 8. Because publications in topic unchanged pattern have higher citation and we consider citation as one of the factors cumulating collaboration quality. Similarly, we also compare the two patterns with isolated cluster, and both of them perform better than isolated cluster. Based on above result, we conclude that although there is no significant difference of two patterns around lower quality part, the topic unchanged pattern is more likely to facilitate outstanding achievements, which reflects advantage of expertise. In addition, it should be noted that although topic unchanged pattern performs better in Fig. 12, it does not means that the UbiComp community is not necessarily promoting risk taking. The result shows us that there is only a small part of scholars focus on a new research field when it was born, which causes less citation than the field have deeper exploration. Therefore we make a suggestion that scholars should bring new ideas and explore them consistently.

Fig. 12
figure12

The quality evaluation of three patterns about research topic

Discussion

Based on the collaboration network constructed by 3,774 published UbiComp papers, we presented an overview of the development of this conference in our work. Compared with some traditional conferences or journals, the 20 years history of UbiComp is not a long time. However, by analyzing the collaboration network, UbiComp has been proved to be a collaborative one with an average node degree of 5.45 and a cluster coefficient of 0.896. It is meaningful to consider what makes an effective or healthy academic community. Furthermore, by combining our experimental results with the overview of the development of the conference, we would like to address a few comments and suggestions from the following three aspects:

  1. 1.

    To begin with, collaborations between scholars benefit to attracting more scholars who are interested in the field, which is well demonstrated in our paper. In Fig. 2, the curve proves that the ratio of new authors each year has an increasing trend, and it has increased rapidly in recent years especially. Other than that, we compare collaboration network of ubiquitous computing with that of three other fields related to computing science and the statistic results are shown in Table 1. It could be found that PG (percentage of giant component) and AD (average degree) of ubiquitous computing are in the middle while CC (clustering coefficient) is a little greater than others, that means ubiquitous computing has higher collaborating trend but more collaboration clusters, e.g., it maintains the diversity of collaboration when the collaboration is strengthened.

  2. 2.

    Secondly, from Figs. 4 and 5, in which we illustrate the networks of collaboration across institutions and nations. It is clearly that institution-collaboration and nation-collaboration are various. During the collaborations of institutions especially between university and corporation, many ideas will be proposed and have a chance to be applied to our daily life and society, which makes the technologies have a practical significance. On the contrary, practical demands are able to push the development of the field.

  3. 3.

    Besides, the histogram in Fig. 9 that we demonstrated the distribution of topic continuity reveals more than half of the authors make exploration of different topics in the different periods, and the quality of collaboration on topic changed pattern is not much bad compared with topic unchanged pattern. Therefore, finding new problems and conduct thorough studies constantly are favorable to bring up creative and meaningful direction of the field. Additionally, there are also some interesting phenomena we find in our research but do not analyze in the paper because the lack of adequate relevant document supporting. For instance, in Sect. 3.1, we have found several well known researchers, and they have made important contributions pushing the field forward. Specifically, regardless of their prominent achievements, we notice that part of links among scholars are driven from supervisor-PhD relations such like Gregory Abowd and Anind Dey, which echoes aforementioned second expansion form originally and inclines to a win–win situation. However, in Sect. 4.2, the supervisor-PhD relation will affect the result of cohesive pattern, because the most participation of PhD and master students last only few years in contrast with dominating professionals and we also regard them as authors with cohesive pattern. Therefore, in the future work, we will clarify the supervisor-PhD relation and remove them from network to attain a more scientific result.

Conclusion

In this article, we first present an overview of collaboration in the field of ubiquitous computing, examining the collaboration evolution and identify several collaboration patterns. We also define the collaboration quality and quantitatively prove the efficiency of four kinds of collaboration patterns which gives us a hint about how to build a strong and productive research network. According to the interesting findings and results of experiments in the paper, we advocate scholars extend collaboration and make research deeply when finding a new exploring direction. In addition, holding the conference in different countries will benefit enlarge influence of UbiComp and attract more and more scholars to construct creative works in the field. To build a strong and productive research network, scholars should not only focus on diversity and sustainability of collaborations, but also pay attention to the appearance of new research directions and conduct deeply explorations of existing research fields. Besides, the analysis methods employed in this work can also be used in other academic conferences or journals, in order to explore the collaboration in corresponding research fields. We hope this work could serve as the footstone and promote further research including field recognition and collaboration recommendation.

Notes

  1. 1.

    http://dblp.uni-trier.de/db/conf/huc/.

  2. 2.

    http://dblp.uni-trier.de/db/conf/pervasive/.

  3. 3.

    https://dblp.uni-trier.de/db/conf/kdd/.

  4. 4.

    https://dblp.uni-trier.de/db/conf/cikm/.

  5. 5.

    https://dblp.uni-trier.de/db/conf/www/.

References

  1. Barabâsi, A.L., Jeong, H., Néda, Z., Ravasz, E., Schubert, A., Vicsek, T.: Evolution of the social network of scientific collaborations. Phys. A Stat. Mech. Appl. 311(3–4), 590–614 (2002)

  2. Bian, J., Xie, M., Topaloglu, U., Hudson, T., Eswaran, H., Hogan, W.: Social network analysis of biomedical research collaboration networks in a ctsa institution. J. Biomed. Inform. 52, 130–140 (2014)

  3. Ch’Ao, L.U., Zhang, C., Shutian, M.A.: How does citing behavior for a scientific article change over time? a preliminary study. Proc. Assoc. Inf. Sci. Technol. 52(1), 1–4 (2016)

  4. Chen, H.H., Gou, L., Zhang, X., Giles, C.L.: Collabseer: a search engine for collaboration discovery, Proceedings of the 11th annual international ACM/IEEE joint conference on Digital libraries, pp. 231–240. ACM (2011)

  5. Coccia, M., Wang, L.: Evolution and convergence of the patterns of international scientific collaboration. Proc. Natl. Acad. Sci. 113(8), 2057–2061 (2016)

  6. Fung, H.N., Wong, C.Y.: Scientific collaboration in indigenous knowledge in context: Insights from publication and co-publication network analysis. Technol. Forecast. Soc. Change 117, 57–69 (2017)

  7. Huynh, T., Hoang, K., Lam, D.: Trend based vertex similarity for academic collaboration recommendation, International Conference on Computational Collective Intelligence, pp. 11–20. Springer (2013)

  8. Huynh, T., Takasu, A., Masada, T., Hoang, K.: Collaborator recommendation for isolated researchers, 2014 28th International Conference on Advanced Information Networking and Applications Workshops, pp. 639–644, IEEE (2014)

  9. Jayabharathy, J., Kanmani, S., Parveen, A.A.: Document clustering and topic discovery based on semantic similarity in scientific literature, IEEE, pp. 425–429 (2011)

  10. Jensen, S., Yu, Y., Liu, H.B., Liu, X.: Query-centric scientific topic evolution extraction, Asis&t Meeting: Information Science with Impact: Research in & for the Community, (2015)

  11. Klemmer, S.R., Thomsen, M., Phelps-Goodman, E., Lee, R., Landay, J.A.: Where do web sites come from?: capturing and interacting with design history, Proceedings of the SIGCHI conference on Human factors in computing systems, 1–8. ACM (2002)

  12. Korada, S.B., Montanari, A., Oh, S.: Gossip pca. Acm Sigmetrics Perform. Eval. Rev. 39(1), 169–180 (2011)

  13. Kronegger, L., Mali, F., Ferligoj, A., Doreian, P.: Collaboration structures in slovenian scientific communities. Scientometrics 90(2), 631–647 (2012)

  14. Liu, Y., Goncalves, J., Ferreira, D., Hosio, S., Kostakos, V.: Identity crisis of ubicomp? mapping 15 years of the field’s development and paradigm change, Acm International Joint Conference on Pervasive & Ubiquitous Computing, (2014)

  15. Lopes, G.R., Moro, M.M., Wives, L.K., De Oliveira, J.P.M.: Collaboration recommendation on academic social networks, International Conference on Conceptual Modeling, pp. 190–199. Springer (2010)

  16. Lu, K., Dietmar, W.: Delineating citation concepts, Asis&t Meeting on Navigating Streams in An Information Ecosystem, (2010)

  17. Luo, H., Niu, C., Shen, R., Ullrich, C.: A collaborative filtering framework based on both local user similarity and global user similarity. Mach. Learn. 72(3), 231–245 (2008)

  18. Masada, T., Takasu, A.: Extraction of topic evolutions from references in scientific articles and its gpu acceleration, (2012)

  19. Newman, M.E.: Scientific collaboration networks. I. Network construction and fundamental results. Phys. Rev. E 64(1), 016131 (2001)

  20. Newman, M.E.: Coauthorship networks and patterns of scientific collaboration. Proc. Natl. Acad. Sci. 101(suppl 1), 5200–5205 (2004)

  21. Shi, X., Leskovec, J., Mcfarland, D.A.: Citing for high impact, Joint Conference on Digital Libraries, (2010)

  22. Sun, Y., Barber, R., Gupta, M., Aggarwal, C.C., Han, J.: Co-author relationship prediction in heterogeneous bibliographic networks, 2011 International Conference on Advances in Social Networks Analysis and Mining, pp. 121–128. IEEE (2011)

  23. White, R.W., Jose, J.M.: A study of topic similarity measures, International Acm Sigir Conference on Research & Development in Information Retrieval, (2004)

  24. Yap, I., Han, T.L., Shen, L., Liu, Y.: Topic detection using mfss. Adv. Appl. Artif. Intell. 4031, 342–352 (2006)

  25. Zeng, J., Cheung, W.K., Li, C.h., Liu, J.: Coauthor network topic models with application to expert finding, 2010 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology, vol. 1, pp. 366–373. IEEE (2010)

  26. Zhang, Q., Feng, Z., Li, X., Zheng, X., Zhang, L.: 25 years of collaborations in ieee intelligent systems. IEEE Intell. Syst. 6, 67–75 (2010)

Download references

Acknowledgements

This work was supported in part by the National Science Fund for Distinguished Young Scholars (No. 61725205), and the National Natural Science Foundation of China (Nos. 61960206008, 61772428, 61972319, 61902320).

Author information

Correspondence to Zhiwen Yu.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Li, Q., Yu, Z., Yi, F. et al. Investigating collaboration in ubiquitous computing research. CCF Trans. Pervasive Comp. Interact. 2, 66–77 (2020). https://doi.org/10.1007/s42486-020-00029-z

Download citation

Keywords

  • Ubiquitous computing
  • Collaboration revolution
  • Collaboration pattern
  • Network analysis