CoPs definition

In Wenger (Reference Wenger1998) CoPs are treated as the informal relations and understandings developed in mutual engagement on an appropriated joint enterprise. In other words, a community of practice is defined as a group that coheres through “mutual engagement” on an “appropriated enterprise”, and creates a common “repertoire”.

In Wenger et al. (Reference Wenger, McDermott and Snyder2002) the concept is redefined towards a more managerial stance, making the concept more popular and simple. Here CoPs are “groups of people who share a concern, a set of problems, or a passion about a topic, and who deepen their knowledge and expertise in this area by interacting on an ongoing basis.” According to Cox (Reference Cox2005), this definition is much vaguer than the previous one. The definition is of a group that is somehow interested in the same thing, not closely tied together in accomplishing a common enterprise. The purpose is specifically to learn and share knowledge, not to get the job done. From this point of view a CoP has three structural features:

1. (i) Domain: it “… creates common ground and a sense of common identity. A well-defined domain legitimises the community by affirming its purpose and value to members and other stakeholders. The domain inspires members to contribute and participate, guides their learning, and gives meaning to their actions” (Wenger et al., Reference Wenger, McDermott and Snyder2002).

2. (ii) Community: it “…creates the social fabric of learning. A strong community fosters interactions and relationships based on mutual respect and trust. It encourages a willingness to share ideas, expose one's ignorance, ask difficult questions, and listen carefully. Community is an important element because learning is a matter of belonging as well as an intellectual process.” (Wenger et al., Reference Wenger, McDermott and Snyder2002).

3. (iii) Practice: it “…is the specific knowledge the community develops, shares and maintains” (Wenger et al., Reference Wenger, McDermott and Snyder2002).

Considering our objective of a systematic approach based on the e-mail exchanged, we are interested in the practice feature of a CoP that is to say the way knowledge is created, shared, and managed.

Different types of CoPs are identified in the literature. McDermott (Reference McDermott2000) indicates four types of community:

1. (i) Communities which are linked to a strategic objective;

2. (ii) Communities which focus on tactical processes, process optimization, and sharing of best practice;

3. (iii) Project-based communities

4. (iv) Communities developing a particular body of knowledge.

We are interested in project-based communities. This kind of CoP has several interesting features. First of all, it is confined in time with fix start and end, enabling to study all its lifecycle phases as detailed in Lee et al. (Reference Lee, Suh and Hong2010) from the building stage corresponding to the initial state of a CoP, where people come together and develop a detailed plan of structure of community to the close stage, during which a CoP declines or transforms into several other communities.

Secondly the people involved in the CoP are project members and therefore act actively in the achievement of the project. Their participation degree, as described in Wenger et al. (Reference Wenger, McDermott and Snyder2002), is a core group. A core group is the group of people that carry out the work in the community. It actively participates in discussions and identifies the topics to be addressed within the community. The coordinator takes place in this group. He/she is the one who organizes events, connects communities and generally keeps the community alive. Moreover, project members are known at the outset enabling to study their mutual interactions easily.

CoP performance evaluation

A CoP is defined as successful when its members exchange specific knowledge, practices and/or experiences that contribute to developing a practice (know-how) in a specific field (McDermott, Reference McDermott2004). The research around the performance evaluation and management of CoPs gains gradually importance. Indeed, in the 90s CoPs were presented as spontaneous, self-organizing, and fluid processes that management cannot intentionally establish (Brown and Duguid, Reference Brown and Duguid1991; Lave and Wenger, Reference Lave and Wenger1991). In other words, at that time it was considered that the performance of a CoP cannot be measured and managed to improve its efficiency. Later some works suggest that CoPs are amenable to manipulation and can be managed (Wenger, Reference Wenger2000; Wenger and Snyder, Reference Wenger and Snyder2000; Lesser and Everest, Reference Lesser and Everest2001). In turn, as a result of these works, diagnosis frameworks for CoP efficiency management were proposed. These aim at assessing the knowledge creation or sharing level inside a given CoP. To do so, as stated in Kim et al. (Reference Kim, Hong and Suh2012), these frameworks are either based on subjective methods or on objective one. Subjective methods are methods that use qualitative constructs to assess the performance level whereas objective one assess this performance level through a quantitiative indicators. In the next sub-sections we detail two diagnosis frameworks one based on a subjective method (Borzillo and Kaminska-Labbe, Reference Borzillo and Kaminska-Labbe2011) and the second on an objective one (Kim et al., Reference Kim, Hong and Suh2012). For each category we analyze its advantages and drawbacks in order to set our research objectives.

CoP diagnosis framework based on subjective methods

In Borzillo and Kaminska-Labbe (Reference Borzillo and Kaminska-Labbe2011) the underlying hypothesis of diagnosis is that organizations need to guide CoPs to generate usable knowledge sustainability. As a result, the authors aim to elucidate the related knowledge creation dynamics. Therefore, they go beyond the mere evaluation of the four factors – knowledge objectives, leadership, collaboration, and boundary spanning – that are generally associated with knowledge creation in CoPs. Contrary to this “fragmented” approach, they propose an “integrated” one enabling to understand how the interactions between these factors lead to dynamic knowledge creation processes. To do so, they exploit the complex adaptative system (CAS) theory that provides an integrative and dynamic framework to understand the interaction patterns in networks of interdependent agents who interact and are bound by their common needs or objectives. As a consequence, the four factors of knowledge creation are translated according to the CAS theory lenses into four constructs.

First, the knowledge objectives factor becomes adaptative tension that drives self-organization and emergence. It emerges from external constraints and corresponds to the energy differential between the system and its environment. The focus is therefore on the definition of the upper and lower bounds of adaptative tension that will define the “region of complexity” inside which a system is able to create a new order (self-organization) and producing new knowledge (emergence). Second, the leadership factor becomes enabling leadership. In this perspective, the managers' role is to design systems in which distributed intelligence can easily emerge. For CoPs, enabling leadership means enhancing the socialization between individuals. Third, collaboration becomes enhanced cooperation. Indeed, CAS theory emphasizes that knowledge creation depends on the nature of the interactions or connections between agents implying for CoPs regular meetings, workshops, and the enabling information technology for interaction. Fourth, the boundary spanning factor remains unchanged. It highlights cognitive diversity for knowledge creation by interacting with actors external to a given CoP.

Then, during 4 years, five CoPs from an international-operating industrial group were studied via a longitudinal exploratory study (six series of interviews) in order to uncover if the four constructs of CAS theory impact positively or negatively the dynamics of knowledge creation. The data gathered were analyzed qualitatively (preliminary analysis for data categorization, within-in community analyses to search linkages between first-order concepts, cross-community analysis to reveal consistencies and contradictions between the CoPs) and quantitatively (average ranking calculation of each constructs). As a result, two modes of CoPs are proposed. Each mode couples two of the four studied constructs (see Fig. 1). When adaptive tension and enabling leadership are prevalent, the CoP is in a “guided” mode. When enhancing cooperation with boundary spanning are prevalent the CoP is in a “self-directed” mode.

Fig. 1.

Guided and self-directed modes of CoPs (Borzillo and Kaminska-Labbe, Reference Borzillo and Kaminska-Labbe2011).

These modes are then coupled with knowledge creation processes. The authors argue that a “guided-mode” supports knowledge expansion, while a “self-directed mode” simulates knowledge probing. The guided mode is used to improve the existing product offerings, which requires creating improved knowledge. During the “self-directed” mode, communities explore radically new knowledge.

This study is interesting as it exploits complexity theory to focus on the interaction of the factors leading to knowledge creation in CoP. Even if this link is not formally described, there is generally acknowledged that there is a positive correlation between performance and knowledge creation. However, it seems difficult, on a given CoP, to evaluate the four constructs: adaptive tension, enabling leadership, enhanced cooperation, and boundary spanning. It requires time-consuming data collection [interviews from different informants (community sponsors, leaders, and members); attending to CoP workshops] and data exploitation. Moreover, only two modes are described in this paper the other possible configurations of the constructs are not considered. That is why another research stream explores the use of objective methods.

CoP diagnosis framework based on objective methods

In Kim et al. (Reference Kim, Hong and Suh2012) a framework diagnosis for CoPs is proposed based on SNA (Social Network Analysis). SNA is a scientific method to analyze a social network by focusing on patterns of relationships between actors and examining the availability of resources and their exchange between actors. Here, performance evaluation of CoPs focuses on knowledge sharing activity by providing a view of the relationship network between the members of the CoP that is to say of the knowledge receivers and the knowledge propagators and the related amount of knowledge exchanged as shown in the conceptual framework of Figure 2.

Fig. 2.

Conceptual framework of knowledge sharing activity in a CoP (Kim et al., Reference Kim, Hong and Suh2012).

The diagnosis methodology proposed by Kim et al. (Reference Kim, Hong and Suh2012) is the following (see Fig. 3):

1. 1. Pre-process: this step enables to understand the methodology of knowledge sharing and to build the knowledge sharing matrix. This matrix records the knowledge propagators in columns, and knowledge receivers in rows. It will be the input for the SNA;

2. 2. Analysis: SNA and development of new indexes for CoP diagnosis;

3. 3. Strategy: suggestion of a strategy for future knowledge sharing activities.

The SNA is based on the knowledge sharing matrix set up during the pre-process step. If the CoP has n members the matrix will be an n × n size one with each cell filled. The input data are generally retrieved from questionnaires, interviews, and data log, even if other transaction data could be used (online messenger tools, e-mails, etc.). The data processed have to define if two members exchange data or not. For e-mails it requires to identify the sender and the recipients of a given message. It is stated that the working out of this matrix spends tremendous time. Moreover, the kind of questionnaires to use is not detailed.

Fig. 3.

Diagnosis process of CoPs based on Social Network Analysis (SNA) (Kim et al., Reference Kim, Hong and Suh2012).

By using a knowledge sharing matrix for input data, basic indexes can be generated by SNA. Some of the basic indexes are:

• • Node type (transmitter or out-flow only node, receiver or in-flow only node, carrier or node with only one connected in-flow, and only one out-flow except from the in-flow node, ordinary or node with a mixed in- and out-flow, isolate or node that is not connected to others);

• • Network density that is an indicator for the general level of connectedness of the graph;

• • Betweenness centrality: it is the share of times that a node i needs a node k (whose centrality is being measured) in order to reach a node j via the shortest path;

• • In and out degree of centrality: it is the proportion of a connected edge to the maximum possible connections.

Based on these indexes the analysis step can be carried out. To do so, the authors provide a member and a CoPs typology. Both typologies are based on knowledge propagating and receiving abilities. There are four kinds of members (see Fig. 4):

• • Balanced player: a member who propagates knowledge to and receives knowledge from other members. This kind of member corresponds to the ordinary or carrier nodes;

• • Egoistic propagator: a member who propagates knowledge to other members, but does not receive knowledge from other members. Such a member corresponds to the transmitter nodes;

• • Egoistic receiver: a member who receives knowledge from other members, but does not propagate knowledge to other members. Such a member corresponds to the egoistic receiver nodes;

• • Knowledge isolator: a member who does not propagate knowledge to or receive knowledge from other members. These members are the isolator nodes.

Based on the member typology the CoP typology is set up (see Fig. 5). It is based on the knowledge receiving and knowledge propagating core group ratio that is to say the ratio between the propagators or the receivers and the total number of member in the core group. The core group is identified thank to in- and out-degree centrality. Four communities' types are proposed:

• • Active community: the core group has a high ratio of knowledge propagation and receiving;

• • Spreading community: the core group has a high ratio of knowledge propagation but a low ratio of knowledge receiving;

• • Learning community: the core group has a low ratio of knowledge propagation but a high ratio of knowledge receiving;

• • Inactive community: this community has low ratios of knowledge propagation and receiving within the core group.

Then, diagnosis of the CoP under study (third step of the framework) is made according to the CoP typology. For each community type improving strategies like, for example, “create more practical knowledge” or “redefine the knowledge domain” are proposed.

Fig. 4.

CoP member typology according to Kim et al. (Reference Kim, Hong and Suh2012).

Fig. 5.

CoP typology according to Kim et al. (Reference Kim, Hong and Suh2012).

This diagnosis framework is complete as it observes a CoP according to knowledge propagating and receiving actions and proposes improving strategies. However, this approach has two main drawbacks. First, the setting up of the knowledge sharing matrix is complex and time-consuming, even if e-mails could be exploited to retrieve the required data automatically. Secondly the diagnosis focuses only on the knowledge exchanges on a binary mode (knowledge receiving yes/no and knowledge propagating yes/no). There is no in-deep analysis of the quality and the type of exchanges that would be interesting to better highlight the CoP performance.

Communication analysis

Several approaches study how to analyze communication as a specific discourse. We note for instance, tagging work in Yelati and Sangal (Reference Yelati and Sangal2011), in which the authors present techniques that help to identify topics in e-mails. We also note NLP (Natural Language Processing) community on automated speech act identification in emails (Baron, Reference Baron1998; Corney et al., Reference Corney, De Vel, Anderson and Mohay2002; Carvalho and Cohen, Reference Carvalho and Cohen2006). For instance, Kalia et al. (Reference Kalia, Motahari Nezhad, Bartolini and Singh2013) use NLP in order to identify messages concerning tasks and commitment. They parse verbs and sentences in order to identify tasks and track messages between senders and receivers. Pragmatics, the study of language in use, is concerned with the intended meaning of speakers beyond what is explicitly stated. It is a branch of linguistics concerned with the use of language in social contexts and the ways in which participants produce and comprehend meanings through language. Pragmatics focuses on aspects of signification that are not only predicted by linguistics knowledge. It is concerned with the analysis of the speaker's meaning rather than on the meaning of words and utterances (semantic or linguistic meaning). Thus Pragmatics takes into account the role of physical and social context (Austin, Reference Austin1975). Pragmatics analysis of communication using e-mails uses only some of these methods like ngrams analysis by Carvalho and Cohen (Reference Carvalho and Cohen2006), Verbal Response Mode scheme by Lampert in Lampert et al. (Reference Lampert, Dale and Paris2010) or a custom coding scheme like in Felice and Deane (Reference Felice and Deane2012).

As cited above, we use Pragmatics in order to study communication in the community of practices and topic identification. Our aim is first to identify if contributors learn from each other's and on which topics. So, we apply the CaMCa “Context aware Mediated Communication analysis” approach (Rauscher et al., Reference Rauscher, Matta and Atifi2016), we develop for this aim. CaMCa is based on Pragmatics analysis and context awareness. So, it helps to identify, from one side, intention of communications and from another side, links interactions to the activity context. CoPs activity evaluation needs these dimensions because participants activities and roles are important. As the analysis is context dependent, it makes no sense to study a big volume of interactions such as in SNA. The objective differs from SNA for e-mails and that could be used to improve the CoP diagnosis framework of Kim et al. (Reference Kim, Hong and Suh2012). Indeed, as detailed in Tang et al. (Reference Tang, Pei and Luk2014) SNA for e-mails is generally used for contact identification focusing on the structure of the network built from the email corpus and paying less attention to email contents.

CaMCa approach

Pragmatics puts on the fact that a dialogue is context and time-dependent. Identifying the sense of interaction is related to the conditions, environment and situation of communication. We develop the CaMca approach that considers from one side the context and the domain of the activity and from the other side the speech acts in mediated communication.

The different phases of this approach are (Rauscher et al., Reference Rauscher, Matta and Atifi2016) (see Fig. 6):

• • Context identification:

  • ◦ Skill and role of actors;

  • ◦ Phases of collaboration among time;

  • ◦ Goal of collaboration.

• • Domain identification

  • ◦ Domain topics and subjects.

• • Communication analysis

  • ◦ Sender/receivers/CC;

  • ◦ Date/Hour;

  • ◦ Subject;

  • ◦ Thread of communication: Reply, comments;

  • ◦ Main Speech Acts.

Each phase of CaMCa will bring a piece of information about CoP activity. Indeed, the context identification is useful to underline the actors' skills and collaboration roles corresponding to the “Community” feature of a CoP as described in the section “CoPs definition” This enables to identify the core group and the coordinator of the CoP under study. The domain identification will be used to identify the topics on which the CoP exchanges, enabling to define the nature of knowledge shared. That is the “Domain” feature of a CoP. Last but not least the Mediated Communication Analysis focuses on the e-mail exchanged enabling to assess the degree of knowledge sharing and interactions between CoP participants (“Practice” feature) and the related knowledge processes that take place.

Fig. 6.

CaMCa basic approach.

In fact, in context we try to identify the organization of the communication. It mainly concerns actors and their collaboration goals. The domain analysis puts on, the nature of the activity subject of interactions, on which practices participants discuss. Finally, the analysis of communication helps to emphasize the real effects of interactions: what happened when actors received and post messages? Is there any learning, coordination, conflict, alliances, etc? And how it is done? CaMCa is used in order to identify the nature of interactions in CoPs. Then the results obtained can be compared to the CoP typology depicted in Figure 4. In this way, the performance and diagnosis of studied CoPs can be put on in a systematic way.

So, we propose to apply CaMCa to a project based CoP. Interaction around a project is analyzed and first principles of systematic CoPs diagnosis approach is determined. These principles are identified by analyzing the communication of the real SEPOLBE project, presented in the next section.

Study context: the SEPOLBE project

We apply the CaMca approach to the SEPOLBE research project in order to identify the knowledge shared and the learning processes that take place between the members and to analyze how the CoP is coordinated (Who is the coordinator?; Level of coordination). The SEPOLBE project is dedicated to develop bioadmixtures for concrete (Goepp et al., Reference Goepp, Munzer and Feugeas2014). These substances are conscientious of the environment and should limit the biocontamination of the concrete surface and improve the resistance to corrosion of its metallic reinforcement. This project implies four research teams and a company. The research teams have different complementary areas of competency: concrete surface analysis, physical chemistry analysis of films on steel and concrete, electrochemistry for steel corrosion inhibition, petrophysics for concrete physical chemistry characterization. Competences in microbiology, chemistry, and microscopy are also required to develop the substances and to analyze the surface bio-contamination. The industrial partner manufactures and markets concrete products such as admixtures. Its product mix already includes protection products but none of them is dedicated to biological contamination. The duration of this project is about 4 years. Actors come from three main domains: Microbiology, Electrochemistry, and Civil engineering. To show the applicability of our approach, we focus on two tasks of the project: (i) Project coordination and (ii) Assessment of the cleaning ability of the mortar base surfaces. Here, the objective is to evaluate the ability of the bioadmixtures to limit the development of biofilms on the concrete surface and its impact on the cleaning concrete surface (biofilm dropping out). The ICube and B2HM teams are in charge of this task. The ICube team has to provide to the B2HM team “adequate” concrete samples. The B2HM is in charge of the contamination and cleaning tests.

For these tasks we analyze the e-mail exchanges between the people involved in the task. The people involved provided us the e-mails they received and sent to complete a given task (coordination or cleaning ability evaluation). So, we had access to the e-mails as a whole (sender, recipients, date, content, etc.)

Learning evaluation using E-mail analysis

The main topics of the task dealing with the assessment of the cleaning ability of the mortar base surfaces concern: Concrete, Mortar, Sample, Bioadmixtures (or BA), Molecular, Bacteria, Essay, Experiment, Polishing, Sample, Ultrasonic, etc.

Actors communicate together using mainly e-mails but if the content of message is available other sources of electronic messages could be exploited such as forums could be exploited. So, we analyze their communication in order to understand if there was any learning of procedures or concepts. First of all, our expert on Pragmatics identifies a grid of main speech act types concerning learning (see Table 1).

Table 1.

Main speech acts related to learning communication

Then, based on different synonyms and sentence forms given by the pragmatics expert and the list of project topics, the NLP algorithm (Lucen) has been used in order to retrieve the corresponding messages. Figure 7 illustrates the global results obtained. The inside wheel shows the topics identified and the outside wheel the corresponding speech acts.

Fig. 7.

Results of NLP analysis.

In our analysis, we try to identify if there is learning of concepts or procedures. So, we look for some specific speech acts like proposition, explanation, request, verification, and information. Then we study this type of interaction among time: at the beginning, in the middle, and at the end of task and project. For instance, at 20/12/12 Charlotte “civil engineer” asks about the Bioadmixtures experiment conditions needed. Thierry “Microbiologist” answers her by explaining a procedure (see Table 2).

Table 2.

Communication between actors for procedure clarification

The text in bold indicates the verb that enables to define the kind of request.

At 21/03/13, some months after this interaction, Charlotte “civil engineer” presents some modification on the procedure (see Table 3).

Table 3.

Modification of the procedure

The text in bold indicates the verb that enables to define the kind of request.

We suppose then that Charlotte learns the procedure, uses it, and tries to adapt it to specific conditions.

At 10/04/13, Charlotte asks for more verification about the samples conservation and Bioadmixtures test conditions of sample treatment. Chao and Thierry define more specifications about these conditions (see Table 4).

Table 4.

Procedure verification about sample conservation

The text in bold indicates the verb that enables to define the kind of request.

Charlotte asks then more verification about the procedure she applies (see Table 5).

Table 5.

Verification of the procedure for sample shipping

The text in bold indicates the verb that enables to define the kind of request.

We suppose that among these interactions, actors learned from each other about bioreceptivity mixture experiment and sample treatment and conservation. Other analysis of messages show also interactions about samples' name coding and sending modes.

The chronology of interactions proves their dynamicity especially if actors have a timeline to respect, they communicate about the project phases.

Coordination analysis

Using CaMca approach, e-mails are also analyzed in order to analyze how the CoP is coordinated. For this purpose, our analysis is based on coordination intentions (Matta et al., Reference Matta, Atifi, Sediri and Sadgal2011). A specific grid was defined containing speech acts related to coordination and topics concerning the coordination (see Table 6). The main coordination speech acts are about information, proposition, and request. Topics are around meetings, documents, and reports. A statistical analysis is also done in order to identify the engagement of actors in interactions. This type of analysis can complete those done by SNA in Kim et al. (Reference Kim, Hong and Suh2012), especially the CoPs users behaviors studied. Adding to statistical studies of exchanges, the identification of the interaction roles as used in our study help to emphasize the dynamic organization movements.

Table 6.

Coordination of main speech acts and topics

In total 42 messages have been analyzed using this grid and based on the CaMCa approach. These messages correspond to four milestones of the projects: Kick off, 12th, 18th, and 30th month. Message date, Senders, Receivers, speech acts, topics are identified. A total of 101 important sentences are identified from these messages. Table 7 illustrates a part of the analysis for the Kick off milestone.

Table 7.

Example of analysis of messages

First statistics analysis synthesized in Table 8 shows that Francoise is the main animator of the group. She takes the role of animator of coordination. She is the author of 22 messages. In these messages, Françoise informs about project documents, meeting reports, and meetings logistics (19 speech acts), proposes meetings schedules and project presentations (17 speech acts) and asks for modification on Meetings Reports, logistics, and project documents. We observe also the dynamic participation of other actors like Bernard (Two speech acts on Information and one on Request about Meeting logistics) and Anouk (one speech act on Information and one Request about Meeting logistics) in the organization of meetings.

Table 8.

Summary of interaction analysis

Finally, even messages are addressed to all participants of projects (16 participants), only seven participate in the discussions. Time interaction analysis shows that messages are close to meetings dates.

As showed in this analysis, CaMca can be used to identify the nature of animation of CoPs and the core group. In SEPOLBE, we can see that some actors are engaged in the animation of CoPs and form the core group that is essential for a successful CoP. The animation is dynamic around meetings and tasks deadline. Besides that, interactions concern explanations of procedures and precision of techniques.