PDS (PlanâDoâSee) cycle. The possibility of the proposed planning support functions is also discussed through flowchart diagrams, and an example based on ...
Planning Support in Capability Growth Trajectory Mining Thongchai Kaewkiriya*1*2 Ryosuke Saga**2 Hiroshi Tsuji**2 Faculty of Information Technology Thai-Nichi Institute of Technology, Thailand* Graduate School of Engineering Osaka Prefecture University, Japan** {thongchai}@tni.ac.th {saga, tsuji}@cs.osakafu-u.ac.jp Abstract This study proposes two functions based on &00, SULQFLSOHV WR JXLGH D VWXGHQW¶V academic growth development plan by evaluating their growth history. One is ³Iinding message generator´ )0* function for dependency capability pairs and the other is ³individuals plan generator´ ,3* function, which is based on a PDS (Plan±Do±See) cycle. The possibility of the proposed planning support functions is also discussed through flowchart diagrams, and an example based on 60 VWXGHQW¶V samples is provided. Keywords: Trajectory mining, Data mining, planning support function.
1 Introduction Data mining is becoming very popular in many areas, such as the analysis of busiQHVV¶JURZWK>1], and is also being applied with personal e-learning systems to assist in online learning and teaching [2]. The consideration of the growth prospects for VWXGHQW¶V growth capability based on trajectory mining is an example of earlier research in which a student capability growth structure is extracted from growth trajectories [3]. The objective is to find growth patterns within student sample groups. Capability structure extraction from growth trajectory to identify constraints on growth and to extract the capability structure of a student dataset has been proposed [4]. In the past, we proposed the formal concept of trajectory mining in the capability space to implement an information system [5]. A practical application involving software outsourcing demonstrated the
proposed concept [1]. It used the basic concept of Capability Maturity Model Integration (CMMI) to evaluate software development on the basis of its capability and maturity [6]. Consideration of maturity and capability expands from initial development to optimization. Capability assessment involves five levels: Level 1 (the lowest) to Level 5 (the highest). In [7], the concept of CMMI was applied to growth trajectory analysis by dividing each capability and application by growth level for science teachers sharing knowledge based on subject contents. However, the research discussed above only proposed the concepts of trajectory mining, such as finding pattern growth, identifying constraints of growth, and applying these to the trajectory and application. Therefore, to continue these concepts, the evaluation of how to apply the technique to real data is required. Learning plans are essential to students because through good plans and recommendations, students can achieve their academic goals more easily and successfully, and can also learn more effectively. To guide students in a study plan, this study proposes a function that generates planning support for students based on their growth history. The intent of this plan is to recommend study areas and assist students in planning their studies. This paper consists of three parts. First, we will present spiral learning on capability space. Second, we will introduce the planning support functions. Third, we will provide an example of planning support for students.
2 Spiral learning on capability space 2.1 Term Definition
ACIS2013 | 81
Let us define first the capability space and capability structure by assigned capabilities Ci: C1,C2,C3«Cn. Each capability Ci has some levels: li1, li2, li3,.., . Generally, each has five levels: li1, li2, li3, li4, li5. Unit Ui represents a growing object such as a student or teacher. For example, a student in English class seeks to improve capability such as listening capability, speaking capability, etc. We can prepare five levels: low, fair, good, very good, and excellent.
capabilities. Furthermore, we define capability subspace as a set of state vectors that exist within two or more capability axes. We take two axes Ci and Cj, for example. Figure 2 shows an example where Xi(t) is a capability state for unit Ui at Xi (1) = (l11, l21, l31 «On1). We assume units grow only by one capability for one change from state (li1, lj1) to (li+1, lj+1).
2.2 Visualization path of growth trajectory A visualization path is a path of each capability level represented by a vector type. Each path can reach a destination or goal depending on inGLYLGXDO¶V current capability level. We can create a visualization path by analyzing data from a student¶V growth log or that from a questionnaire if log data is unavailable or for a new student.
Figure 2. Two-axis based Capability Space A capability structure suggests dependency between two . For example, high level such as li5 in Figure 3(a) can be achieved without any pre-condition imposed by Cj or vice versa. However, achieving level li5 as in Figure 3(b) may sometimes require satisfying another condition. In this example, li5 is achieved only if a unit achieves li4 first, but li4 requires that li3 and lj3 are met first, thus li3, lj3, and li4 are prerequisites to achieving li5.
Figure 1. Example of growth trajectory As shown in Figure 1, the growth trajectory indicates the path that can be followed to increase a capability by visualization of the path and interrelations between capabilities. In the example, Ci represents listening capability and Cj is reading capability. There are five achievable levels (poor, average, good, great, and perfect). Three possible paths are shown, indicating ways to improve their capability levels in the two capabilities. We now define the capability space. The basic idea comes from evaluation of knowledge space theory [8], [9] for knowledge acquisition. Then, capability space is a set of potential
(a)
(b)
Figure 3. Example of capability structure 2.3 Plan±Do±See Cycle Plan±Do±See is a principle applied to improve the quality of general management [10] (Figure 4 ³3ODQ´is preliminary analysis and prediction of effects that may occur on account of upcoming
ACIS2013 | 82
changes. For example, ³3lan´ is to pass the Japanese language proficiency N1 test within five \HDUV ³'R´ LV what is being done and there is little to change in plan control. For example, a diligent student reads and practices Japanese all the time with the intent of passing the N1 exam as SODQQHG ³6HH´ LV WKH HYDOXDWLRQ ZRUN WKDt has already been done and revised to improve the VWXGHQW¶VFKDQFHVRIPHHWLQJWKHJRDO. For example, a student has done as planned, then it is determined whether the N1 has been passed or not. If the goal has not yet been achieved, the plan is revised. Thus, planning support is based on the Plan±Do±See cycle.
Table I Comparison of functions
Planning support function will support ³SODQ´ DQG³6HH´EHFDXVHWKHSURFHVV RI H[Wracting the planning support function must have to plan from student growth log. Then, we will get the extracted data as the planning result (See Figure 7). )RU ³6HH´ WKH UHVXOW RI H[tracted planning support function will show the result from the past until the future which students have been recording. This means the follow up and evaluation from planning.
3 Planning support function 3.1 Basic principles Figure 4. Plan±Do±See cycle 2.4 Review of previous systems The previous research [5] presented the development of a prototype system for supporting growth by using visualization of the growth history. The research was divided into three main parts: first was the input of capability data for students. For this part a dataset (learning history data) was imported into the system. The second part was to visualize the learning trajectory based on the learning history data from the first part. Visualization included two types: individual and group visualization. The last part was the analysis of history data in order to recommend learning paths to a student. This system would require extensive manual intervention in order to analyze the results for multiple students, thereby requiring a large amount of analysis time and cost. Moreover, the system only supports WKH ³6ee´ part of the Plan±Do±See cycle. Table I shows the comparison between available functions of the process discussed in [5] and this research. Visualization function support supported only the evaluation purpose (See). FMG (finding message generator) planning shall support WKH ³3ODQ´ function. The proposed planning support function supportVERWK³3ODQ´DQG³6HH´ functions in the PDS cycle.
In this study, we use the basic principles of CMMI that are used to evaluate software organization through analysis of software capability and maturity. Both maturity and capability are indicators of consistency and growth trends reflecting the quality of a software product. CMMI applies growth levels to maturity and capability from level 1 to level 5, as shown in Figure 5.
Figure 5. Capability Maturity Model Integration In addition to CMMI, we introduce the Spiral Enhancement Capability Support (SPICE) system [11], as shown in Figure 6. SPICE is a learning support system that can store and update student growth logs. In this paper, we use SPICE to store and analyze logs for growth trajectory and capability structure planning. SPICE also allows users to share experiences with each other. Moreover, SPICE allows a supervisor to improve student capabilities through effective planning.
ACIS2013 | 83
Section 3.4. Finally, the last step provides overall results. One result comes from the FMG function to recommend enrolling students in the types of student groups they will benefit from. The result from the IPG function will recommend an individual study plan. 3.3 Flowchart for generating a message Figure 6. Model of SPICE 3.2 Process for planning support function
Figure 7. Finding message process Let us divide the process of the planning support function into four steps. The first process is to record student histories. This process collects data through student questionnaires in some cases (if documented history data is not available). Students provide achievement data per year reflecting their current capability levels. The second process is to build a visualization path for each capability level. Each path can achieve the planned destination depending on an LQGLYLGXDO¶Vcurrent capability level. The third step involves generating the planning support function. This step consists of two functions. (1) FMG is the extraction message from student learning histories. The algorithm used to generate messages is Interpretive Structural Modeling (ISM) [12]. This section focuses on the finding message for dependencies between capability pairs. The details of the finding message for dependency capability pairs will be discussed in the next section. (2) Individual Planning Generator (IPG) is the plan generated from extraction RIDVWXGHQW¶Vlearning history. The process of extraction is explained in
Figure 8. Flowchart for generating a message (Step f1). FMG follows three steps as follows: Step f1, (Figure 8) we collect the values of Relation and Necessary by setting all capability pairs in the object class. For this process, we use Loop (Legacy Calculation Loop) to calculate the values of Necessary. Then, we create Object class, which is the agent for Relation, and then collect DOO5HODWLRQV¶GDWD$[LVL$[LVM1Hcessary) in that page to create a list used in Step f2. Step f2, (Figure 9) Searching for the Strong Dependency Relation and no Dependency Relation. Searching for Strong Dependency: As the first step in searching for Strong Dependency values, we define the variable Max Relation Count for determining the maximum number of relationships of type Relation. An array will be packed
ACIS2013 | 84
Figure 9. Flowchart for generating a message (Step f2). only with Strong Dependency Relation values (the Strong Dependency Relation Array). We begin by setting Max Relation Count equal to 1. After that, we perform the Loop to evaluate each Relation obtained from the previous step. We determined that the value of relationships of Relation in the Loop is higher than that of Max Relation Count, and then perform the following three steps: 1) Increase Max Relation Count to equal to the number of relationships of Relation in the Loop. 2) Clear the Strong Dependency Relation Array. 3) Insert the Relation value in the Loop into the Strong Dependency Relation Array. ,I WKH YDOXH RI 5HODWLRQ¶V UHODWLRQVKLS LQ WKH Loop is equal to Max Relation Count, it inserts the Relation value in the Loop into the Strong Dependency Relation Array. If the value of ReODWLRQ¶V UHODWLRQVKLS LQ WKH /RRS LV ORZHU WKDQ Max Relation Count, no action is taken. Searching for No Dependency: When searching for No Dependency, we apply a principle similar to that for searching for Strong Dependency, except that, in this case, we determine whether the value of each ReODWLRQVKLS¶s Relation is equal or not equal to 0. If the number is equal to 0, we place the Relation into the No Dependency Relation Array.
Step f3 simply displays the Strong Dependency and No Dependency. We apply the Strong Dependency Relation Array and No Dependency Relation Array as display data. 3.4 Flowchart generator
for
individual
planning
We describe the flowchart for the IPG function in this section. The process begins by reading output data from the visualization function. Then, we set advice or recommendation type to create planning support. Later, the student status is checked. The process has three conditions that must be checked; the first condition checks whether a student is over-focused on one side of the capability structure, and if so, the result will recommend that the student should get back to studying other skills first. The second condition determines if a student can or has completed two FDSDELOLWLHV LQ OHYHO WKH UHVXOW LV ³