Process recommendation technologies have gained more and more attention in the field of intelligent business process modeling to assist the process modeling. However, most of the existing technologies only use the process structure analysis and do not take the social features of processes into account, while the process modeling is complex and comprehensive in most situations. This paper studies the feasibility of social network research technologies on process recommendation and builds a social network system of processes based on the features similarities. Then, three process matching degree measurements are presented and the system implementation is discussed subsequently. Finally, experimental evaluations and future works are introduced.
Workflow technology has gained further application and development with the rapid growth of modern business environment. And process models are being widely used in the development of organizational structures [
This paper is organized as follows. After this introduction and the related works in Section
A process model is often in form of some graphical notations and describes how a certain process is composed out of different tasks, in which resources are involved in carrying out these tasks and objects are manipulated [
Social network has been widely researched in the last decade and various social network systems have been built, some of which also produce great commercial success. Social network analysis (SNA) has become an important branch of scientific studies. The related studies cover a wide range of different fields, including sociology, psychology, economy, and computer science [
From the discussion above, the greatest contribution of paper [
Business process is a series of activities that are performed by different performers to specific targets, respectively, and the order of activities represents collaboration of these performers. In workflow systems, the process reflects the actual business process, and the activity node represents business operation in enterprise. Information or operations will flow or be conducted in turn according to the nodes sequence (or by the arcs). Therefore, the business process model can be abstracted as a directed graph, of which the nodes stand for activities and the arcs stand for orders.
Let
In a workflow net,
Next, we give a formal definition of social network as the following.
A social network, SN, is a graph denoted by a 5-tuple
As the main manifestation of social network, SNS (social network software/social network site) has gained huge success in business and become the most concerned topic in related research fields. In [
A more recent definition is proposed by Kwon and Wen [
SNS is more and more applied or studied in business domain and also attracts research attention as marketing instruments [
Almost in all cases, a process has four main structures, including AND-split, OR-split, and AND-join, OR-join. In short, the type “
Good and bad constructions.
Given that every process is verified not to have bad constructions before it is added into the process repository, it is enough to consider the good constructions only. Therefore, we can only consider the AND-split or OR-split and the AND-join or OR-join will be implied in the good constructions.
For simplicity’s sake, the places and transitions can be combined as nodes, and the tasks, performers, split type, join type, and other properties of places and transitions can be added into the nodes’ properties. Under this idea, business process is a series of activities that is performed by different performers to specific targets, respectively, and the order of activities represents collaboration of these performers. In a workflow system, the process reflects the actual business process, and the activity node represents business operation in enterprises. Information or operations will flow or be conducted in turn according to the nodes sequence. Therefore, the business process model can be abstracted as the directed graph, of which the nodes stand for activities, with a corresponding label (implicating activity type, content, the serial number, etc.), and the edges stand for orders.
Let
Node
However, in a social network site, for example, twitter or Sina Weibo, there exist several interexchange methods, such as tweet, @, reply, and retweet. Tweet is the conduct of publishing a message, retweet is to forward a message that others tweeted without any change, @ is used to push the message to others by force, and reply tends to accept message self-selectively with no or more additional information. Thus, when every performer in a social network site is a user, their relationships that are implied in process structure can be regarded as information exchange in the social network. For example, if there is a node “
It is important to note that the users being performers in the same process only depend on whether they @ or reply the same topic (the message series) when reconstructing a process from social network. But what is the first message in a message series? We define a message series code for each message series and the code is also the only representative of the response process. So the message series code can also be called as process code. The first message in a message series is the series code that is published by the process creator. So a whole process can be found from its creator in the social network. In actual applications, a user can search for suitable process models and generate a new model by combining them; for example, user “U1” created a process P1 and tweeted the process code C1, and user “U2” created a process P2 and tweeted the process code C2. When user “U3” creates a new process P3, he may use the P1 and P2 as parts of the P3. Then in the social network, “U3” must retweet C1 and C2 in turn before he tweets the process code C3 and the C3 must contain the C1 and C2. Next, the performer of the first node of P3 will reply the message C3, …, and so forth. Thus, study on social network (2) from user history can be transformed to relationship mining from users that @ or reply the same or related topic. Likewise, study on social network (3) from insertion history can be transformed to relationship mining from users that retweet related topics.
Finally, three types of social networks defined in [
DFS (depth first search) and BFS (breadth first search) are widely used in graph mining algorithms. Compared with BFS, DFS consumes less memory but runs slower due to the stack utilization. On the contrary, the BFS uses more memory but runs faster than DFS. The facts in construction practice of process repositories and process patterns show that the vast majority of business processes are simply structural and with less nodes (in process repository used in this paper, there are 86% processes containing less than 20 nodes, and of which nearly 71% processes contain less than 10). Thus, the problem of memory usage is no longer an issue. On the contrary, because the number of processes may be huge, the time performance is more important. Therefore, this paper uses BFS to define the process code.
Breadth first search by different node orders of the same hierarchy may lead to different BFS sequences. Therefore, this paper presents the standard BFS sequence to ensure that the BFS sequence of the same process is unique. And we can reconstruct unique process with the standard BFS sequence. Next, we give the definitions about BFS code based on BFS sequence.
The breadth first traversal order on a process graph
For example, as shown in Figure
Process sample
The BFS code of process
Except for (
Next, we give the definition of process node label that is used in the process code.
A process node label is a value of label function
The process code is composed of each node’s label, while the message that a user replies or @ comprises the relevant functions, events, and connector.
Note that if a process is a fragment of the other process, then the first process matches the other, but the opposite is not always true. However, for simplicity, we do not consider the difference and regard them matching with each other.
In different intention, process matching degree is related to user matching degree, structural matching degree, and behavioral matching degree.
User matching degree between two processes can be measured based on their creators, performers, and reusers, and thus the user matching degree can be divided into three parts: creator matching degree, performer matching degree, and reuser matching degree. Let
Let
Let
The process user matching degree is a combined metric that can be represented as the following:
The process structural matching degree can be calculated according to process matching degree that is based on BFS code [
A social network schema from social network model is
A metapath of information propagation is a path defined on the network schema
For example, in the process sample
Label of role.
Role |
|
|
|
|
---|---|---|---|---|
Label |
|
|
|
|
As we discussed above, a process may have one or more metapaths in the social network. The matching degree of the two metapaths can be measured by the SED [
Given two strings
Secondly, we give the definition of process structural matching degree.
Given two processes
The behavioral matching degree mainly measures the dynamic features about functions and events. Generally, functions and events are named by the unified naming conventions (if not, it can use the ETL tool and semantic analysis to standardize the names). Given processes
Social network supported process recommender system turns to help process builders fitting processes to achieve a modeling intention with regard to the other builders’ modeling intention and modeling behavior.
This recommendation system implementation consists of three parts (as shown in Figure A simple social network site meets our research needs by providing the basic functions of tweeting, retweeting, @ing, replying and searching, and so forth. A process import method is also provided in this part and each process, in the form of a well-formed XML document, can be added into the SNS database automatically. A query interface allows users to request process models or process model parts that are of interest to them. In the system, we use open-source MySQL database to provide index capability, but not well in large dataset task. Therefore, we use Lucene to build indices on the objects of processes and network and provide query on creators, performers, reusers, functions, events, connectors, and so on through the query parser syntax. A series of recommender components proposes appropriate process models which fit to a business process model that is currently being edited under different considerations. The parameters configuration is also provided for the users to determine the execution sequence of three types of process matching degrees and the adjustment coefficients for the process user matching degree. The thresholds of three matching degrees in certain recommending proceeding can be set through parameter configuration.
Structure of recommender prototype.
Different parameters values may lead to different efficiencies, even different results. In the next section, the difference will be demonstrated and discussed.
In this section, a comprehensive study is conducted in our experiments on synthetic dataset. The synthetic data generator that we use is similar to [
In the experiment, 5 process fragments (identified as (1), (2), (3), (4), and (5) in Figure
Experiment performance.
The average execution time and the average number of recommended processes are listed in Table
Experiment result.
Parameters configuration | Average time (ms) | The average number of recommended processes | |
---|---|---|---|
(1) |
|
43 | 5 |
|
|||
|
|||
(2) |
|
76 | 5 |
|
|||
|
|||
(3) |
|
51 | 5 |
|
|||
|
|||
(4) |
|
84 | 8 |
|
|||
|
|||
(5) |
|
55 | 11 |
|
|||
|
|||
(6) |
|
53 | 17 |
|
|||
|
In the same execution order, adjusting every threshold in turns may lead to big change in the results. From Table
We propose a social network supported process recommender and related techniques which can improve process modeling by providing reference processes from different perspectives to extend or complete the business process under construction, especially in some social considerations.
The similarities between process features and social network features are discussed in detail. And three types of process matching degrees between two processes are proposed by the consideration of contexts in different situations; then we implement the system prototype. The experimental evaluation shows that our method is efficient and effective for practical use.
The greatest contribution of this paper is that we show how to add social features to a recommendation-based process modeling support system. And the unique social network constructed from the process repository is first proposed which can assist the researchers in understanding social features in process modeling. The three types of process matching degrees are the initial attempt that benefits from the social network. At a more abstract level, this paper strengthens a research stream into process modeling that combines social software and social computing.
Still, much work has to be carried out in the future. For example, the user community with the similar interest may be more important for the modeling intention of process. Furthermore, process mining can benefit from the social network analysis. We hope that this paper can inspire more researchers to join in the newly ascendant field and many more efficient and effective methods or technologies can come to the fore.
The authors declare that there is no conflict of interests regarding the publication of this article.
This work is supported by the National Natural Science Foundation of China under Grant (no. 61272129), the National Science and Technology Supporting Program of China (no. 2012BAH06F02), the Research Foundation for the Doctoral Program by Ministry of Education of China (no. 20110101110066), the New-Century Excellent Talents Program by Ministry of Education of China (no. NCET-12-0491), and the Zhejiang Provincial Natural Science Foundation of China (no. LR13F020002).