Developing Teams by Visualizing Their Communication Structures in Online Meetings

Abstract

This research pursues the question of how the computer-generated analysis and visualization of communication can foster collaboration in teams that work together online. The audio data of regular online video meetings of three different teams were analyzed. Structural information regarding their communication was visualized in a communication report, and then, discussed with the teams in so-called digitally supported coaching (DSC) sessions. The aim of the DSC is to improve team collaboration by discerning helpful and less helpful patterns in the teams’ communication. This report allows us to recognize individual positions within the teams, as well as communication structures, such as conversational turn taking, that are relevant for group intelligence, as other research has shown. The findings pertaining to the team members during the DSC were gathered via questionnaires. These qualitative data were then matched with the quantitative data derived from the calls, particularly social network analysis (SNA). The SNA was inferred using the average number of interactions between the participants as measured in the calls. The qualitative findings of the teams were then cross-checked with the quantitative analysis. As a result, the assessment of team members’ roles was highly coherent with the SNA. Furthermore, all teams managed to derive concrete measures for improving their collaboration based on the reflection in the DSC.

1. Introduction

1.1. Purpose and Approach of This Research

Supporting online collaboration becomes more and more relevant as communication has significantly shifted from personal to virtual meetings in the wake of the COVID-19 pandemic. This has affected professional, educational, and private life. The tool and the method of digitally supported coaching introduced here aims to improve team online collaboration. It does so by analyzing and visualizing patterns of communication, such as participants’ interactions as well as the social structure of the team, reflected by the relative positions of team members. This approach is based on the experience that visualizing communication patterns and structures can be helpful in understanding team dynamics, as explored in the field of sociometry [1] The visualization of social interplay to support learning is also a common concept in some team coaching approaches [2]. The approach presented in this paper provides a seamless and efficient technique of capturing data through extensions of the video conferencing tool Zoom™. The generated set of visualization is integrated into a digitally supported coaching (DSC) scenario. The current paper describes all components of the DSC along with three case studies aimed to validate our approach and envisage its further development.

From the viewpoint of Watzlawick’s communication theory, understanding the patterns of communication is key to understanding the underlying social system: “The search for pattern is the basis for all scientific investigation. Where there is pattern there is significance—this epistemological maxim also holds for the study of human interaction” [3]. From this theoretical perspective, it seems worthwhile to use the possibilities of computer analysis to help detect and visualize patterns in online communication to support the reflection of team collaboration. The approach examined in this analysis seeks to make teams more aware of the unwritten rules of their communication by helping them to discern constructive and less constructive patterns in their communication.

To this avail, the author has developed a tool that analyses recorded online video meetings and visualizes the communication that took place. Once three to four meetings of the same kind with the same team have been visualized and compiled into a communication report, this report is discussed with the team as part of a digitally supported coaching session. The objective of the DSC is for the team to gain insight into their communication patterns and to learn how to conduct their meetings more constructively. Conversational turn taking can be viewed in the interaction and airtime diagrams; the “rhythm” of the conversation, for instance, whether there are long monologues or dense interactions, can be seen in the communication sequence diagrams, as Figure 1 illustrates. Qualitative information gathered right after each meeting provides insight into which meetings participants perceived as more constructive or efficient than others.

As illustrated in Figure 1, the interaction diagram shows the number of interactions between the meeting participants. The interaction data are also used to conduct a social network analysis (SNA) between the players, with each node representing one team member, and the ties between actors representing the interaction between these team members. The thickness of the tie represents the number of interactions. This allows us to compute the centrality metric of each team member. The quantitative data form a basis for cross-checking the internal subjective views of the team members in the DSC, thus providing different views on team communication: model-based quantitative data and the perceptions of the team members.

This research investigates to what extent teams can learn from this approach employing a multiple-case study. The focus is on two aspects of online team collaboration: the roles and the interaction within teams. Other aspects of communication and team coaching could be pursued using this approach in further research.

1.2. Theoretical Foundation

Watzlawick’s communication theory is the foundation for understanding the term communication pattern and its significance for team communication used in this research. Watzlawick uses the game of chess as a simile for human communication [3]. An outside observer, who does not know the game of chess, could deduct the rules by watching a series of games played. Communication, when seen as interaction between players and the inherent lawfulness of this interaction, is defined as a stochastic process in this theory. “Stochastic process refers to the lawfulness inherent of symbols or events”, for instance, “the patterns of tonal and orchestral elements by a composer”. A stochastic process implies that in each context, some events are more likely to happen than others. In Mozart’s piano sonatas, an andante is likely to be followed by an allegro. When a certain team approaches a decision, it might be that many remain silent while there is intense staccato-like interaction by two or three people followed by a certain person having the last word. For the purpose of this research, Watzlawick’s concepts are followed in defining communication patterns as the inherent lawfulness of interaction within a certain group. It is a lawfulness that we tend to be oblivious to, as Watzlawick argues: “we are particularly unaware of the rules being followed in successful, and broken in disturbed, communication”. It is a lawfulness that can be brought to light by observing the communication of several meetings, as the first proposition of this research states: discerning communication patterns in a team can help the team to improve their communication.

Gregory Bateson introduced the idea of metapatterns, which he called symmetrical and complementary dynamics. A team where members strongly compete would be an example of a symmetrical communication dynamic. An example of a complementary dynamic would be team members taking different roles that complement each other. Bateson argues that either pattern, when taken to the extreme, would lead to the breakdown of the social system: “On the one hand, various sorts of symmetrical rivalry among individuals were observed and it was evident that such rivalrous sequences of interaction could be progressive [i.e., self- reinforcing] and therefore ultimately pathogenic” [4].

Reflecting communication patterns can also be key to improving group intelligence. As empirical research has shown, the equality of conversational turn taking was the largest factor in predicting group intelligence [5]. Group intelligence in this research was defined by the performance groups showed in work-related tasks like brainstorming, judgement, and planning. Patterns of conversational turn taking are a type of communication pattern that can be measured by analyzing the interaction between participants without analyzing the content of what has been spoken.

The team coaching approach followed in the DSC is based on constructivist–systemic coaching. This approach “encourages incremental changes in terms of small steps toward the desired position on a particular topic” in seeking to “facilitate movement toward desired team positions as defined by the team” [6]. Constructivist–systemic coaching has roots in constructivist therapy [7]. One basic assumption of constructivist therapy is that “reality arises from consensual linguistic processes” [8]. In this view, the distinction between what is helpful and what is detrimental to good team communication emerges during the team discussion when reflecting the communication report. This is incorporated in the second research proposition: studying their communication patterns can help team members to learn about their roles and their interactions within a team.

All findings in these team coaching processes are specific to the context of the respective team. This follows another fundamental assumption of De Shazer, that “the meaning of words is determined by how they are used by the various participants within a specific context”. De Shazer shows an understanding of language that is rooted in Ludwig Wittgenstein’s philosophy as expressed in his later work [9]. Therefore, which communication patterns are deduced from a communication report and whether they are considered helpful is highly context-specific.

2. Related Work

Extensive work exists on the relevance of communication patterns in therapy and coaching. Here, “questions are formulated to bring forth the ‘patterns that connect’ persons, objects, actions, perceptions, ideas, feelings, events, beliefs, contexts, etc., in recurrent or cybernetic circuits” [10].

Sociograms and the value of visualizing social systems were widely explored by J.L. Moreno [11]. His initial empirical study of a New York School, including the methods employed there, has become a reference for sociometry until today and has been further elaborated on [1].

Samrose and colleagues evaluated ways of analyzing and visualizing online video communication [12]. As early as 2003, Anne Massey analyzed and visualized temporal patterns in global virtual project teams to explore “the nature of team interaction and the role of temporal coordination in asynchronously communicating GVPTs” [13]. In this research, temporal patterns are defined as the flow of interactions over time that influence communicative, decisional, and interpersonal behaviors in teams. Other research has covered the graphical analysis of indicators “which measure the intensity of various aspects of users’ communication activity in periodic manner” [14]. Here, aspects such as variances in the project activity or differences in users’ activity levels were visualized for the identification of various types of team communication issues, for instance, when a “user only asks for information but does not provide any”. Research has supported that in distributed online team work, “synchronous communication encourages social interaction and expressions of different views” [15].

Various pieces of work exist on the use of social network analysis. As in this research, some work used SNA as part of a mixed-method approach. Mixed-method approaches can be defined “as research in which the investigator collects and analyzes data, integrates the findings, and draws inferences using both qualitative and quantitative approaches or methods in a single study or a program of inquiry” [16]. The use of SNA in combination with critical discourse analysis (CDA) in a mixed-method approach was used to explore the social dynamics in a classroom [17]. Here, CDA and SNA were combined to detect the change in informal leadership in a group of pupils in a classroom from one person to another over time. Another mixed-method study was conducted to explore social systems, aimed at discerning so called guanxi circles in Chinese organizations. Guanxi circles are “pseudofamilies in a person’s working life and at work usually develop from ego-centered social networks around one focal person” [18]. A mixed-method approach combining SNA and Natural Language Processing was used to analyze Massive Open Online Courses (MOOCs). [19]. Text obtained from the MOOC was analyzed and mapped to social presence using the theoretical framework of Shea et al. [20], thus creating information regarding the social position of the students. This was then cross-checked with the results of the SNA, particularly regarding the in-degree centrality and the authority score.

Research on sociological phenomena that involved digitally measuring human interactions was carried out by Alex Pentland, who called this branch of research social physics [21]. In his research, Pentland, for instance, found that the “rhythm” of interaction of the highest-performing teams consisted of “many short contributions rather than a few long ones” [21]. Pentland further stated that “in studies of more than two dozens of organizations I have found that interaction patterns typically account for almost half of all the performance variation between high- and low-performing groups” [21].

3. Research Questions and Propositions

This research pursues the following questions for teams that collaborate online through video meetings:

• RQ1: What can teams learn about their communication from reflecting upon its visualization in interaction diagrams?
• RQ1.1: How well can the individual positions of team members be recognized by the team through this approach?
• RQ1.2: How well can the interplay within the team be recognized by the team through this approach?

This research is based on the following propositions:

• Discerning communication patterns in a team can help the team to improve their communication.
• Studying a team’s communication patterns can help team members to learn about their roles and their interactions within a team.
• Visualizing patterns is a helpful way to convey them to the team.

4. Methods

4.1. Research Design

This research is designed as a multiple-case study following theoretical replication logic as outlined by Yin [22]: the same research questions are pursued with the same research design for three selected cases. Two of the cases are IT companies of similar size operating in the same geographical region. One organization is a large international company.

Each case is bound by encompassing a specific group in a specific organizational context. The sessions recorded are all part of the same online meeting series serving a specific function for the respective teams, e.g., a weekly jour fixe. The meeting invitees of all cases remained constant in the period observed, the actual participants per meeting varied in some instances due to sick leave or other absences.

To increase the construct validity of the study, each case has two sources of evidence: quantitative and qualitative data.

4.2. Quantitative Data

Quantitative data were obtained through the meeting recordings. The meetings were conducted using the video conferencing tool Zoom^TM (https://zoom.us, accessed on 30 October 2021, Version 5.7). This tool was chosen as it allowed us to store separate audio files for each participant when recording the meeting. From these audio files, the so-called diarization of each speaker was derived automatically through the software written for this research, including the point in time and length of each time this participant spoke in the meeting. Further (aggregated) data were then automatically calculated by combining the diarization data of all speakers:

• The “airtime”: the percentage of speaking time each speaker had overall.
• The number of times each specific speaker spoke after another speaker (interaction).
• The sequence of speaking events throughout a meeting.

Point 2, the interaction, serves as the basis for the social network analysis. Interaction is regarded as one person speaking after another person. It is based on the assumption that in most cases, the statement of a person represents an “Anschlusskommunikation”—a communication that is connected to what has been said before. This creates a margin of error for communications in which statements are not related to one another and therefore the communication does not represent an interaction. In non-pathological communications, as researched here, we assume that this is the exception, not the rule, and therefore does not interfere with the search for communication patterns.

Based on this interaction diagram, a social network analysis was conducted. “The objective of social network analysis is the description and analysis of relationships between different actors” [17]. In this study an actor is defined as one person within the team. Actors are labeled as P₁ … P_n. “Actors are connected by ties that are determined, for example, by behavioral or physical interaction” [17]. In this study, one tie represents the number of times a person spoke after the other in one specific meeting, and t₁₁ is the number of times A spoke after B in meeting 1.

A relation differs from a tie as it reflects a social aspect between two actors that is relevant to the entire network. It is “the measurement of different ties in an overall network” [17]. Relations in this study reflect the interaction between two actors over the course of all observed meetings. They are weighted as follows: the higher the number of interactions, the higher the weight of the relation. Accordingly, the relation between P₁ and P₂ is defined as:

$$R_{12}=\frac{\sum _{i=1}^n{t}_{i1}+t_{i2}}{n}$$

where n is the number of meetings that both P₁ and P₂ attended.

Based on this, the degree centrality of each team member can be calculated. As the social network is constructed using a graph with weighted nodes (with the relation R as described above), the information of the weights can be taken into account when calculating the centrality measure. As a suitable way to compute degree centrality for social networks with weighted nodes, the following definition was chosen [23]:

$$s_i=\sum _{j=1}^m{w}_{ij}$$

where m is the total number of nodes, w is the weighted adjacency matrix, in which w_ij is greater than 0 if node i is connected to node j, and the value represents the weight of the tie.

Various alternate definitions for centrality in social networks with weighted nodes have been proposed. The centrality definition of “the number of nodes that a focal node is connected to, and the average weight to these nodes adjusted by the tuning parameter [23], for instance, takes both the weight of the nodes as well as the number of relations an actor has into account. This can be advantageous when we have a social network with a very uneven number of relations between its actors. However, this comes at a price: this metric requires a parameter alpha to balance the relevance of node weight and the number of relations. This increases the complexity of the metric and requires empirical work to justify which value to use for alpha in what context. Another approach is to view role positions based on the number of incoming ties, seeing a prestigious actor as one to whom many ties are directed [24]. This forms a concept that has been carried on, for instance, to define centrality in asymmetric networks [25]. Other centrality measures cater to complex social networks with subgroups [26].

In the current study, which investigates small social networks, the advantages of more complex metrics do not seem to outweigh their disadvantages. Following the principle of Occam’s razor, it seems prudent to apply the least complex model and increase the complexity of the model as more complex problems (in this case, more complex social networks) are being investigated using this approach.

The quantitative data are also used to create visual representations of the communication in each meeting. This process is supported by the software, with the interaction diagram currently being the only one needing manual composition utilizing the automatically generated data. Together, these visual representations compose a communication report encompassing the following:

• The airtime, visualized as a pie chart, as illustrated in Figure 1;
• The number of times each speaker spoke after another speaker, visualized as an interaction diagram;
• The sequence of speaking events throughout a meeting, visualized as a sequence diagram.

4.3. Qualitative Data

Qualitative input data were collected at a digitally supported coaching (DSC) session. At this event, the team members met to reflect their communication based on the communication report of their meeting series. The DSC was facilitated by an experienced constructivist–systemic coach.

Upon reflecting on the communication report, the team could exchange different viewpoints on the overall communication structure. Furthermore, everybody could check their own feeling against the real data on how involved they were. Communication patterns could be discerned when viewing the visualization of several meetings side by side. The team members could then ask themselves which of their patterns were helpful and what might be worth changing to attain more constructive communication. This reflection process was structured in the DSC following a uniform agenda.

The DSC followed the same predefined agenda:

• Introduction.
• Going through the report (communication visualizations) of one meeting.
• Team members are asked to reflect on:
  • The whole “image” that they see in the interaction diagram and the communication sequence diagram.
  • Their own communication behavior that they recognize in the visualizations.
  • Their particular communication behavior, and what (if anything) surprised them. Furthermore, they were encouraged to share what they derived from what they saw.
• Giving an overview of all recorded meetings: airtimes and interaction diagrams.
• Looking for recurring patterns, following the question “what is typical for our team communication?”.
• Answering of the questionnaire by all meeting participants.
• Feedback on the report.
• Closing.

In point 6 of the DSC, the participants filled out the following questionnaire:

• What communication patterns did you detect through this report? (things that are typical for your communication)
• What contributed to constructive communication?
• What could you as a team change to achieve more constructive communication?
• What contributed to efficient communication?
• What could you as a team change to achieve more efficient communication?

The DSC questionnaire was filled out by all DSC participants in cases 1 and 2. In case 3, only the moderator found the time to fill out the questionnaire. This case is nevertheless used in this study as its other data are fully available and provide relevant findings.

These qualitative data reflect what insights were obtained by the participants during the digitally supported coaching. Researching question 1, the results of the questionnaire are used to see what teams can learn about their communication through this team coaching approach. The sentences of the questionnaire served as coding units. They were analyzed through a reflexive thematic analysis as outlined by Braun and Clarke [27]. Codes were formed inductively as recommended by Mayring [28], and recurring themes were identified.

4.4. Mixing the Methods

In this research, qualitative and quantitative data represent two sources of evidence. Each reflects the communication that took place in the meeting series in a different way. The social network analysis represents the metrics that can be inferred from the actual speaking times. The qualitative information obtained from the questionnaires is used to find out what the participants learned from the DSC as defined in research questions 1.1 and 1.2. These data represent the subjective perspectives of the participants: what they see and consider noteworthy in the visualizations of the communication they find in the report after having discussed it with the other team members.

To answer research questions 1.1 and 1.2 the qualitative data were cross-checked with the metrics of the social network analysis (which had not been disclosed in the DSC) to see what similarities and differences the qualitative findings and the quantitative metrics hold. Regarding RQ 1.1, the degree centrality, as a quantitative measure to assess the centrality of a person with the team, was compared with what the team members expressed in the questionnaires. This way, it could be assessed to what extent the insights team members gained regarding their role within the team matched with the SNA. Regarding RQ 1.2, the measure of the weight of relations was compared with the internal view of the team regarding the intensity of their interactions. This way, the results of the qualitative analysis regarding team interactions could be compared with the results of the quantitative analysis.

4.5. Participant Recruitment

The organizations participating in this research were chosen because they had previous contact with the author in his work as organizational consultants or through academic cooperation with the author or the author’s colleagues. All organizations who were willing and had resources to participate at that point in time were included in this research. Before starting the research, all members of the participating teams were informed about the research and how the data would be anonymously used for research purposes. They were then asked for their consent to participate under these terms. All members of the three teams agreed.

5. Results

5.1. Case 1

5.1.1. Team and Meeting Series

The team in case 1 consists of four team members, two men and two women. They meet every week for one hour online in their weekly jour fixe. The jour fixe is of the operative kind, supported by quantitative data as input prepared prior to the meeting. The team is part of a company of around 100 people. Team member P2 could not participate in the DSC; therefore, answers to the questionnaire are only available from P1, P3, and P4.

5.1.2. Analysis of the Roles and Interactions within the Team

The team viewed the communication report, including the airtime and interaction diagrams of four meetings beside one another, as shown in Figure 2. This graphic is specifically designed so that teams can detect patterns in their communication more easily. After the reflection of the communication report, the team members regarded P4 as the person leading the meeting, as the qualitative analysis shown in

Table 1. Results of the thematic analysis of the three questions in the questionnaire in case 1.

What Communication Patterns Did You Detect through This Report?	What Contributed to Constructive Communication?	What Could You as a Team Change to Achieve More Constructive Communication?
P4 leads the meeting	Clear roles	Pay attention to topics other than one’s own
Meetings are structured	Communication was balanced and at eye-level	Give the team more time to add topics
Interaction mainly takes place between P1 and P4	Everybody was involved (even if the parts have a different weighting)	P3 could be more involved in the meeting
Interaction mainly runs through P4	P4 dispatching to all others	More interaction between participants apart from P1
I hold back in the meeting

indicates. This is coherent with the social network analysis, where P4 turns out to have the highest degree centrality, as

Table 2. Degree centralities in case 1.

Team Member	Degree Centrality
P1	255.58
P2	169.67
P3	90.33
P4	330.75

shows. One theme of the analysis stated that interaction mainly ran through the meeting leader, and another theme more specifically pointed out that the highest interaction was between P1 and P4. This is coherent with the SNA, where the most intense relation is between P4 and P1, followed by P4 and P2, as

Table 3. Relations and their weights in case 1.

Relation	Weight of Relation
P1–P2	52.33
P1–P3	33.75
P1–P4	169.50
P2–P3	6.33
P2–P4	111.00
P3–P4	50.25

shows. The participants nonetheless regarded their communication as “balanced and at eye-level”, as the qualitative analysis shows.

The team further reflected on the strong centrality of their leader, when asked what could contribute to achieving more constructive communication. One theme stated that “More interaction between participants apart from P1” would be helpful, and another found that P3 could be more involved. This reflects the current situation, where P3 has the lowest airtime and the lowest degree centrality, as the SNA shows.

5.2. Case 2

5.2.1. Team and Meeting Series

As in case 1, the team in case 2 consists of four team members, two men and two women. They meet every week for one hour online in their weekly jour fixe. The jour fixe is of a more general kind; input is sometimes provided, but is not a meeting prerequisite. The team is part of a company of around 40 people.

5.2.2. Analysis of the Roles and Interactions within the Team

The team, after viewing and reflecting the interactions as shown in Figure 3, identified P3 as moderator and the person with the highest number of interactions, as the result of the qualitative analysis shows (see

Table 4. Result of the qualitative analysis of the questionnaires in the digitally supported coaching session in case 2.

What Communication Patterns Did You Detect through This Report?	What Contributed to Constructive Communication?	What Could You as a Team Change to Achieve more Constructive Communication?
P2 constantly has the fewest, but very balanced and constant, interactions with others	Culture of open and honest communication	Increase meeting efficiency
P3 is visibly the moderator, having the highest airtime and interactions with others	Moderator, who involves everybody	Involve P2 more
The communication reflects who has which task intersections with whom	The meeting frame	Ask good questions
P1 acts like a hub between different people with different areas of responsibility	The regularity of the meeting	More fact orientation
Patterns of the meetings become visible	Good meeting focus
Intensity of interactions between P1 and P4 remain constant across the meetings	Fact orientation
Partial interruptions in communication
Occasional long discussions

). This is coherent with the SNA indicating that P3 has the highest centrality measure, as shown in

Table 5. Degree centralities in case 2.

Team Member	Degree Centrality
P1	462.75
P2	184.75
P3	564.00
P4	492.50

.

P2 was recognized as the person with the lowest speaking time, but the team also recognized that she continuously had quite balanced interactions with all other participants throughout the meetings. Both impressions are supported by the quantitative data: P2 has only 43% of the average number of interactions within the team but keeps her interaction on a similar level to all team members, with a coefficient of variation of 22%. Accordingly, P2 has very similar weights in her relationships with all the other participants, as

Table 6. Relations and their weights in case 2.

Relation	Weight of Relation
P1–P2	48.75
P1–P3	235.25
P1–P4	178.75
P2–P3	75.50
P2–P4	6.,50
P3–P4	253.25

shows.

P1 is regarded as the “communication hub”, having a high relative airtime and a high number of interactions with all other participants. This corresponds with the average actual airtime of 30.34% of the spoken time within these four meetings. The number of interactions between P1 and the other team members is second only to the moderator.

The team also recognized that their communication behavior reflects “who has which task intersections with whom”. In their reflection, the team appreciates a lot of its qualities, such as the communication culture and the value of the moderator. Suggestions for room for improvement are both on an organizational level (increasing meeting efficiency) as well as on a relationship level (involve P2 more).

5.3. Case 3

5.3.1. Team and Meeting Series

Case 3 covers a team of middle management, their supervisor, who is part of the senior management, and the moderator. Altogether, they are 19 people: 10 women (including the moderator) and 9 men. However, not all of them can attend their weekly jour fixe of one hour. The average number of attendees was 11 in the three meetings observed. Their jour fixe is structured into phases featuring a check-in, information and decision phases, and a check-out. The meeting is supported by a tool in which all issues to be discussed are entered prior to the meeting. The management team is part of a large international corporation.

5.3.2. Analysis of the Roles and Interactions within the Team

Here, only the perspective of the moderator could be captured in the qualitative analysis. She clearly discerns several patterns in the communication of this team: She states that “the boss always had the highest airtime”, as

Table 7. Result of the qualitative analysis of the questionnaires in the digitally supported coaching session in case 3.

What Communication Patterns Did You Detect through This Report?	What Contributed to Constructive Communication?	What Could You as a Team Change to Achieve more Constructive Communication?
Airtimes were quite unbalanced	Humor	As moderator, taking more care to have balanced airtimes
Boss always had the highest airtime	Check-in	Better including people who came late to the meeting
Late-comers had low interactions	Strict time management
Strong interaction between moderator and boss	Tool support to manage variety of tasks
Communication ran mainly through moderator

shows. And indeed, the supervisor (P10), who attends this regular meeting with his employees, always had between 37 and 47 percent of the airtime, as Figure 4 illustrates. Secondly, she recognizes, that the highest interaction is between the moderator and the supervisor. Both of these statements are supported by the quantitative data: the highest number of interactions is between the boss and the moderator (with the exception of meeting three, where P7 co-moderated with P1), as

Table 8. Strongest 10 relations and their weights in case 3.

Relation	Weight of Relation
P1–P10	102
P1–P11	66
P8–P10	45
P1–P7	37
P1–P2	34
P1–P17	33
P8–P9	23
P1–P8	22
P8–P13	22
P8–P15	21

shows. Thirdly, she discerns, that communication ran mainly through the moderator. This is also supported by the quantitative data; the moderator has the highest centrality, as Table 8 shows. In fact, there were hardly any interactions other than with the moderator. Only the supervisor occasionally had interactions with other persons. All of the top 10 weighted relations involved either the moderator or the supervisor.

Remarkably, all three aspects of their communication appear when the moderation role is taken by another person; in the second meeting, P8, a stand-in for P1 (who was on vacation at that time), served as moderator. And the airtime of the boss was again by far the highest (47% in a meeting with 12 participants); the interaction was again highest between the moderator and supervisor, and the interactions from all participants were mostly with the supervisor and the moderator. This pattern becomes quite visible when viewing the interaction diagram illustrated in Figure 4. It resembles a star topology of a network with the moderator serving as the hub. This is also reflected by the degree centralities: by far the highest centrality was held by the moderator (P1), as

Table 9. Degree centralities in case 3.

Team Member	Degree Centrality
P1	400
P2	85
P3	55
P4	9
P5	51
P6	25
P7	78
P8	207
P9	64
P10	252
P11	119
P12	42
P13	74
P14	17
P15	33
P16	15
P17	63
P18	44
P19	7

shows. She had a degree centrality of 400, with the median of all centralities being 55. According to the SNA, by far her strongest relation was with the supervisor, being more than 50% stronger than the second strongest relation. (As there were 19 persons in these calls with 171 relations, only the 10 relations with the highest values are shown in Table 8.)

Through the communication report, the moderator also recognized that people who were late to the meeting and missed the check-in were hardly involved. The moderator concluded that she would strive to ensure more balanced involvement of meeting participants and to better integrate those who are late.

6. Discussion

6.1. General Observations Regarding the Cases

Cases 1 and 2 are comparable insofar as they both feature a team of four persons having a weekly jour fixe meeting of one hour. However, their communication could hardly be more different. As the data derived using the tool shows, team 1 has one dominant speaker with more than 50% speaking time in every meeting. Speaking time in team 2 is more balanced, with an overall number of interactions seven times as high as in team 1. In light of the empirical research on group intelligence [5], it may be asked what the effect of this kind of communication is on leveraging the potential of team 1 compared to team 2. This, however, must consider the nature of the meeting. Group intelligence as measured by Wolley et al. focused on brainstorming, judgement, and planning tasks. There might be meetings where a one-to-many communication pattern, as in team 1, is useful.

The third case is special, as it also shows that communication patterns in a team can prevail even when the moderation role switches from one person to another, as occurred in meeting 2, with the supervisor having most of the airtime and a moderator through whom the communication runs. The most interaction occurs between moderator and supervisor, and together, they capture around two thirds of the airtime of each meeting. If others in the meeting speak, this occurs almost exclusively as a result of the interaction with this duo—a pattern that became overt through the communication report. The communication report of case 3 also unmasked that in this meeting series, several people hardly participated at all. This could be used to raise follow-up questions: What are these people contributing to these meetings? Is this in line with the purpose of this meeting series, and if not, what could be changed?

6.2. Findings Addressing the Research Questions

Regarding RQ1.1, “How well can the individual positions of team members be recognized by the team through this approach?”, the teams reliably found the leader of the meeting, and identified people who were central to their communication and those less involved in cases 1 and 2. Here, the results of the questionnaires matched the analysis of the SNA. In case 3, the dominant role of the boss in these meetings and how the moderator supported that communication pattern became visible.

Regarding RQ1.2, “How well can the interplay within the team be recognized by the team through this approach?”, the team in case 1 quicky discerned that most interaction was carries out through their leader (P4), with a strong interplay between him and P1. P3 was seen as less involved, and she considered herself as “careful to interfere”. All of this is coherent with the centrality measures and the weight of the relations. In case 2, the team also recognized that there was a leader and a second person who increased the interaction. However, the interaction was more balanced and much higher than in case 1 (by 600%). The team also found that one person was less involved in the interaction, but her interactions with others were balanced. Again, this is coherent with the weights of the relations.

Regarding the overarching RQ1, “What can teams learn about their communication from a mixed-method approach using social network analysis and qualitative methods?”, the qualitative data show that the discussion of the communication report led to meaningful discussions in the DSC and to concrete ideas about how to make their communication more constructive in all three cases. The findings for improvements ranged from social aspects, like bringing a certain person more into play, striving for more balanced airtimes as a moderator, and better including late-comers into the meeting, to practical issues like preparing an agenda prior to the meeting or focusing on topics that are relevant to all participants.

This is an initial indication that team members were able to accurately assess their roles and discern patterns in their interactions and learn from the DSC, as proposed by this research.

7. Limitations

A limitation of this study is that it only covered three cases. Another limitation might be that all teams functioned in a Western-dominated culture. Furthermore, the reports were analyzed by one person only, possibly being biased. This potential bias was minimized by discussing the qualitative findings with other professionals in the field.

8. Conclusions

In all three cases examined in this research, the visualizations “spoke” to team members, as they became more aware of any regularities in their communication behavior. They identified aspects of their roles and their interactions based on the communication report that were coherent with the quantitative data. Furthermore, in each case, discussing the communication report helped the teams to discern which response pattern turned out to be constructive and which were less so. This indicates that—under proper conditions—the digitally supported coaching approach as described in this work can contribute to improving the collaboration of teams working together online.

It appears that having a structural view on communication can be a good basis for reflecting the collaboration within a team. The current study provides initial evidence on the applicability and promotive effects of the developed tools, visualization, and digitally supported coaching session structure. Further research across different application contexts and cultures is needed to confirm or relativize the findings presented in this study.

In this research, the focus was on roles and interactions within teams. Other areas of research could focus on how a DSC can lead to more efficient meetings or can help to resolve tensions within a team. Patterns indicating or leading to innovative processes or products might be explored by adapting the approach presented in this paper. Other lines of research could use the quantitative data generated by the tool presented here to explore predictors for group intelligence in teams working together online. A further path would be to use the visualizations and SNA metrics generated by the tool to monitor the integration of new members in a team working together online over time. Moreover, it could be investigated whether predictive machine learning models could be trained to indicate upcoming turning points in conversations, such as conflicts, based on the diarization data generated using this tool. This would follow the assumption that the “rhythm” of a conversation changes in a certain way before a turning point in a conversation occurs.