ASHYI-EDU: Applying Dynamic Adaptive Planning in a Virtual

Learning Environment

Jaime Pavlich-Mariscal

, Yolima Uribe

, Luisa Fernanda Barrera León

, Nadia Alejandra

Mejia-Molina

, Angela Carrillo-Ramos

, Alexandra Pomares Quimbaya

, Monica Brijaldo

, Martha

Sabogal

, Rosa Maria Vicari

and Hervé Martin

Departamento de Ingeniería de Sistemas, Pontiﬁcia Universidad Javeriana, Bogotá, Colombia

Departamento de Informática Teórica, Instituto de Informática, Universidade Federal do Rio Grande do Sul, Porto Alegre,

RS, Brasil

Laboratoire d’ Informatique de Grenoble, Grenoble, France

Keywords:

Adaptation, Personalization, Virtual Learning Environment, Education.

Abstract:

Activity planning is an essential element in the teaching-learning process, since it can ensure that adequate

activities are utilized to convey the information to students. The common practice in course planning is that

the teacher selects the same set of activities for every student in the classroom. However this does not address

the students’ heterogeneity in learning styles, knowledge, and personality. To address this problem, this paper

proposes ASHYI-EDU, a Virtual Learning Environment (VLE) with dynamic adaptive planning. ASHYI-

EDU is able to capture distinctive student characteristics and provide students with plans that are especially

tailored for their particular characteristics. This paper also presents an ongoing case study that utilizes ASHYI-

EDU in a university course.

1 INTRODUCTION

Formal educational processes require several ele-

ments to foster student competencies and abilities.

One key element is the plan of activities that students

need to perform in a course. This plan should satisfy

two main requirements: its activities must be aligned

with the course goals and they have to be aligned with

the students characteristics, so that students can ef-

fectively learn the course contents (Sangineto et al.,

2008).

To create a course plan, teachers usually develop

a unique plan of activities that must be followed by

every student in the classroom. Although this is a

widely utilized practice, it may not guarantee that

the courses’ learning goals are satisﬁed by every stu-

dent. The reason is that students are heterogeneous;

they have different personalities and learning styles,

which means that activities should be aligned with

these characteristics to properly convey the informa-

tion to students. In addition, students have different

degrees of competencies, and abilities, which means

that some students may require additional remedial

activities to properly accomplish the course goals.

Overall the common way of planning a course

may not be the optimal way to ensure effective learn-

ing in students. To address this issue this paper

presents ASHYI-EDU, a Virtual Learning Environ-

ment that assists teachers to create especially-tailored

plans for every student, taking into account their spe-

ciﬁc personalities (Yang and Chen, 2013), learning

styles (Mohamad et al., 2013), competences (Ry-

tikova and Boicu, 2014a), and abilities (Li et al.,

2008a). The word "ASHYI" from ASHYI-EDU is a

Quechua word that mean to search, to investigate. The

combination with "EDU" conveys the idea of search-

ing, investigating to provide education.

Although there are other approaches to create

adapted plans in educational environments (Li et al.,

2008b; Bahmani et al., 2011; Rytikova and Boicu,

2014b), their planning process takes into account

either context, student personality, learning styles,

learning context, but not all of them simultaneously.

Some of them focus on coarse-grained learning ele-

ments, such as courses or learning units, but not ﬁne

grained activities. In addition, some approaches only

rank learning activities but they do not elaborate an

adapted plan for each student.

Pavlich-Mariscal J., Uribe Y., Barrera León L., Mejia-Molina N., Carrillo-Ramos A., Pomares Quimbaya A., Brijaldo M., Sabogal M., Vicari R. and Martin

H..

ASHYI-EDU: Applying Dynamic Adaptive Planning in a Virtual Learning Environment.

DOI: 10.5220/0005430400520063

In Proceedings of the 7th International Conference on Computer Supported Education (CSEDU-2015), pages 52-63

ISBN: 978-989-758-107-6

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

The main contributions of ASHYI-EDU are: a

more detailed analysis of of user and context informa-

tion to perform dynamic adaptive planning and taking

into account the student evolution during the course

execution. Student evolution is relevant, since stu-

dents may learn new abilities or competences dur-

ing the course. As students changes their attributes,

the plans currently assigned to them may not be the

optimal. To address this situation, student plans in

ASHYI-EDU are dynamic, which means that when-

ever students change their characteristics, ASHYI-

EDU creates a new plan that is better suited to these

new student conditions.

ASHYI-EDU also is able to determine whether a

student requires a remedial activity or not, to com-

pensate the lack of certain competences or abilities

that the student requires to complete certain course

activities. Moreover, ASHYI-EDU not only is able to

select the best suitable activities for each student; it

is also able to select the best resources available for a

speciﬁc activity, based on context (e.g. access device,

location).

The remainder of this paper describes ASHYI-

EDU and an ongoing case study to validate the ap-

proach. Section 2 discusses similar approaches for

activity planning. Section 3 brieﬂy explains ASHYI,

the high-level architecture that is the base for ASHYI-

EDU. Section 4 explains ASHYI-EDU in detail. Sec-

tion 5 describes the case study to validate the ap-

proach. Finally, section 6 concludes and discusses

future work.

2 RELATED WORK

Automatic planning is a topic widely addressed in

the literature. This section analyzes three main cate-

gories: generic planning, planning oriented to speciﬁc

context, and approaches for planning in personalized

education.

Table 1 summarizes the main features of related

work. The ‘+’ symbol means that the feature is sup-

ported by an approach; ‘-’ means that it is not sup-

ported, while ‘?’ means it is not clear whether the

feature is supported or not.

Several works address the planning problem with

a general-purpose approach. The work of (Coles

et al., 2010) (1) propose a tool to create plans, based

on partial order planning withing the framework of

forward search to reduce the ordering constraints that

may arise when creating a plan or sequence of activ-

ities. The plan can be modiﬁed over time. The work

of (AnousouyaDevi et al., 2012) (2) presents an ap-

proach to address non-predicable actions using a tem-

poral planning algorithm. The approach of (Plaku

and Hager, 2010) plan collision-free trajectories for

robots, based on the STRIPS algorithm (Fikes and

Nilsson, 1972). The work of (Geib and Goldman,

2009) is PHATT, a tool creates a model that takes into

account the negative experience of the user in the exe-

cution of a plan. The approach of (Geib and Goldman,

2009) (3) creates a stochastic model for the execution

of a plan, based on AND/OR, negative experience and

multi-goal agents.

In regard of automated planning to different do-

mains, the work of (Oh et al., 2009) (4) addresses

meeting planning, focusing on maximizing satisfac-

tion of participants. This planner takes into account

what happens before, during, and after the event. This

approach also calculates the time required by each

activity to minimize wait time between them. The

work of (Guoqi et al., 2010) (5) focuses on task de-

composition as the main concern to plan activities

in a multi-agent environment. This decomposition

takes into account time and order relations between

tasks to identify and resolve conﬂicts between agents.

The approach of (Johnson and Sieber, 2009) (6) ad-

dresses planning for tourists that travel through dif-

ferent places, to determine the impact of external pro-

cesses and speciﬁc planning decisions. This approach

takes into account socio-economical and contextual

elements, such as weather, recycling, etc. The work

of (Marki et al., 2012) (7) present a planning approach

that takes into account the motivation of the user to

execute upcoming activities based on their results in

previous activities and the discomfort originated from

their performance. The approach of (Nijland et al.,

2012) (8) propose a planner for people’s daily agenda.

In this case, most activities are planned in advance,

so planning, re-planning, and programming are per-

formed in different times. The planner takes into ac-

count a person’s data (gender, age, income, education,

occupation, etc.) and creates a plan that also includes

activities that are not explicit in the agenda, but that

may be of importance to the user.

Several works address planning in the educational

context. The work done by (Rytikova and Boicu,

2014b) (9) creates a course from resources provided

by the teacher. The course is presented to each stu-

dent depending on their course competences, which

are obtained through an initial test. The approach of

(Bahmani et al., 2011) (10) propose a tool to recom-

mend courses to students, based on some student at-

tributes and context. Several agents interact to ﬁnd

the best study plan for a given student.

Even though these approaches contribute in differ-

ent contexts and with different methods to the prob-

lem of activity planning, when they are used to plan

ASHYI-EDU:ApplyingDynamicAdaptivePlanninginaVirtualLearningEnvironment

student activities they have some limitations. For in-

stance, they do not facilitate the inclusion of new vari-

ables, such as student personality, learning styles or

learning context, to increase the precision of the ac-

tivity assignment. Besides, some proposals do not

include mechanisms to dynamically change the plan

according to the evolution of the individual. Finally,

some of these approaches do not establish methods

to increase the possibility of success using remedial

activities. All of the above issues are directly ad-

dressed by ASHYI-EDU. ASHYI-EDU has compre-

hensive proﬁle information about activities and stu-

dents to better determine the most alike activities to

students. ASHYI-EDU is able to dynamically adjust

the plan, based on the student evolution and context

changes, and it is able to include remedial activities,

if the student needs them during the execution of the

plan.

Table 1: Related Work.

Approach Activ.

Decmp.

Re-

planning

Agents User’s

satisf.

Context User

data

(1) + ? - - + ?

(2) + ? - - + ?

(3) + ? + + + +

(4) + + - + + -

(5) + ? + - - -

(6) ? + - - + -

(7) + - - + + +

(8) ? + - - + +

(9) + - - + + +

(10) - ? + - + +

3 THE ASHYI ARCHITECTURE

This section brieﬂy describes ASHYI, a high-level

architecture to perform adaptive planning that is the

base of ASHYI-EDU. For more details about ASHYI,

the reader can refer to (Jaime Pavlich-Mariscal,

2015). The main elements of ASHYI are: roles, activ-

ities, resources, context, users, and a generic planning

process.

The main roles in ASHYI are: the Administra-

tor, who characterizes the application domain, deﬁnes

the main activities that can be performed and the user

characteristics; the Planner, who guides the planning

process and solves conﬂicts that the system may not

be able to solve; and, the Executor, who performs the

activities in a plan that the system creates according

to his/her speciﬁc characteristics and context.

Activities are the main actions that can be per-

formed in a plan. They form a hierarchical struc-

ture of composite activities (contain sub-activities)

and atomic activities (have no sub-activities). Activi-

ties may have pre and post conditions that determine

the requisites to execute them and the effects they pro-

duce in the environment, respectively.

Resources are physical or logical elements that are

employed to perform certain activities. Among other

attributes, resources have keywords to identify them

and a speciﬁcation of the way to locate them (e.g.

URLs for logical resources).

Context includes all the external elements that

may affect the interaction between users and the sys-

tem. The main identiﬁed elements are: location, time

(e.g. season, time of the day), location infrastructure

(e.g. computer availability, Internet connection, etc.),

communities (i.e. groups of persons to which the user

belongs), environment (e.g. weather, luminosity, am-

bient noise, etc.), regulations (i.e. rules, laws, stan-

dards that affect the individual), access devices (e.g.

hardware, software, etc).

The user proﬁle characterizes the people who uti-

lize the system. The main information in a proﬁle

are: demographic data, preferences (for different ac-

tivities, information display, etc.), interests (things the

user wants to do), predilections (perceptions that the

user likes), and habits (actions the user performs fol-

lowing a pattern).

The planning process in ASHYI,comprises three

main stages: (1) environment conﬁguration, (2) activ-

ities planning, and (3) activities execution. Stage (1)

deﬁnes the speciﬁc domain characteristics to perform

the planning process. In particular this stage deﬁnes

the speciﬁc resources, activities and their pre and post

conditions that will be utilized in the adapted plans.

Stage (2) utilizes user, and context proﬁles to de-

termines the sequence, dependencies, and conﬂict res-

olution between activities in a meta-plan, i.e., a struc-

ture that describes all of the possible plans that could

be assigned to users. According to the speciﬁc user

and context information, the system must select a sub-

set of the meta-plan that is the best suitable plan for

each user, i.e., the adapted plan.

Stage (3) is the process in which the user executes,

step by step, his/her adapted plan. The user executes

each activity in turn. The results are evaluated by the

system to determine the user’s performance. The sys-

tem may determine that the user has changed and be-

cause of that, the user may need a better adapted plan.

In that scenario, the system should automatically cre-

ate a new adapted plan for that user.

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

4 ASHYI-EDU

To validate ASHYI, the framework was instantiated

into a speciﬁc domain (education). The result is

ASHYI-EDU, a dynamic adaptive planning system

for virtual learning environments. ASHYI-EDU is a

Java-based application that enhances Sakai (Apereo-

Foundation, 2014), a virtual learning environment

software, to provide personalized education to stu-

dents.

ASHYI-EDU addresses several issues in the

teaching-learning process. First, teachers need to

develop plans that convey knowledge, abilities, and

competencies to students. Those plans comprise sev-

eral learning activities that students must perform to

reach speciﬁc learning goals in a course. Since stu-

dents are heterogeneous, they may beneﬁt more from

some type of activities and less from others.

For instance, some students prefer to learn by ex-

ecuting practical activities, while others would learn

better by doing theory-intensive activities. Some stu-

dents may prefer to learn by visual means (e.g. dia-

grams, images, etc.), while others may prefer to ob-

tain the information in textual means.

Consequently, to provide the best learning path to

each student requires that the teacher creates different

plans, each of them providing the best activities for

each student, depending on their speciﬁc characteris-

tics and preferences.

In practice, this is a complex task. A teacher

would have to: (1) Understand the speciﬁc character-

istics and preferences of each student, (2) determine

the best activities for each student depending on the

information of (1) to create a custom plan for each

student, and (3) monitor the execution of each custom

plan. Without the help of technologies, the above task

would be very time consuming for teachers.

The above issues are the motivation to develop

ASHYI-EDU. ASHYI-EDU provides a set of tests

to assess student characteristics and preferences. It

characterizes activities, based on their suitability for

different student attributes and automatically creates

plans with activities that are the most suitable for the

students’ learning processes.

ASHYI-EDU is able to recognize changes in the

student attributes, usually yielded by the successful

learning of speciﬁc abilities or competences. Using

this information, ASHYI-EDU can automatically cre-

ate a new plan that is more suitable to the student.

Moreover, the system is also able to provide reme-

dial activities to the student if he/she needs them to

successfully complete the rest of the activities in the

plan.

Figure 1 summarizes ASHYI-EDU planning and

(1) Configure virtual learning

environment

(3) Course planning

(5) Complete tests

(7) Create adapted plan

(9) Activity execution

(11) Update student profile

(2) Course structure, activities,

resources

(4) Meta-plan

(6) Student profile

(8) Adapted plan

(10) Activity results and feedback

(12) Updated student profile

Figure 1: ASHYI-EDU planning and execution.

execution. The ﬁrst activity is to (1) conﬁgure the vir-

tual learning environment, in which the teacher spec-

iﬁes the course structure and all the possible activities

and resources that can be utilized to fulﬁll those goals

(2). The second activity is the course planning itself

(3), in which the system automatically selects activi-

ties and creates a meta-plan (4), i.e., a structured de-

scription of all the possible adapted plans that could

be assigned to students. When students are going to

interact with the course, they have to complete per-

sonality and learning style tests (5), which provide the

information for their proﬁles (6). The system auto-

matically assigns (7) students a custom plan (8) that

is best suited for their characteristics. The last activity

is the execution (9), in which the student performs the

assigned activities, the teacher evaluates the results of

the activities (10) and updates the student proﬁle (11).

If the student proﬁle (12) changes, the system auto-

matically creates a new plan (7).

The rest of this section describes each of the above

elements in detail.

4.1 Conﬁgure Virtual Learning

Environment

This activity is performed before beginning any

course in the system, and focuses on creating all of

the essential course elements: learning units, learning

goals, activities, and resources.

ASHYI-EDU assists teachers in creating the

above elements by providing the basic data structures

to specify learning units and learning goals, and pro-

viding a searchable repository of activities that the

teachers must ﬁll before initiating any course.

4.2 Course Planning

After the basic course structure is created and a suf-

ﬁciently large pool of activities has been created, the

teacher must perform two main activities: preselect-

ing activities and resources and creating a meta-plan

ASHYI-EDU:ApplyingDynamicAdaptivePlanninginaVirtualLearningEnvironment

rep-

resen-

tative

Agent

Context

Agent

Interme-

diary

Agent

Router

Agent

Informa-

tion

Source

Agent

Context

changes

Query

Select and send

query, based on

type

Collected

results

Query

Select and send

results

Figure 2: Overview of PUMAS-LITE.

that describes all of the possible plans that could be

executed by students. Both activities are described as

follows.

4.2.1 Activity and Resources Selection

One of the key elements in course planning is the

selection of all of the activities and resources that

are aligned with the course goals and learning units.

ASHYI-EDU provides a repository of activities and

resources that teachers can ﬁll to utilize in one or more

courses. Since the repository can be shared by multi-

ple users, other teachers may contribute with activities

and resources that can be aligned with the learning

goals of the current course. To properly preselect the

available activities and resources, ASHYI-EDU pro-

vides PUMAS-LITE, a java-based subsystem that uti-

lizes agents to search the repository according to dif-

ferent criteria: keywords, learning goals, among oth-

ers.

PUMAS-LITE is a lightweight version of

PUMAS (Carrillo-Ramos et al., 2007), a multi-agent

system for knowledge management and adapted

information retrieval with a focus on ubiquitous

environments. PUMAS provides four multi-agent

systems that support the location of resources in an

ubiquitous system: a multi-agent system in the client

device, another one on the server-side that performs

adaptation and search, a third one that acts as a

middleware between the two, and the last one that

performs adaptation.

PUMAS was instantiated into the educational do-

main as PUMAS-LITE, focusing only on the server-

side multi-agent system. Figure 2 is an overview of

PUMAS-LITE. There are ﬁve agents:

Context Agent, which is aware of the user execu-

tion environment and reacts to its changes. There is

one Context Agent per user.

Representative Agent, which represents the stu-

dent within PUMAS-LITE and manages the student

proﬁle. There is one Representative Agent per user.

Intermediary Agent that is in charge of receiv-

ing and collecting queries from different representa-

tive agents, selects those queries and pass them to

a Router Agent. There can be several Intermediary

Agents in the same platform.

Router Agent that is in charge of receiving

queries from the Intermediary Agent and searching

for an information source that could totally or par-

tially satisfy the query. Once the source is found,

the router agent communicates with the Information

Source Agent to retrieve the answer to the query.

Information Source Agent, which contains the

information required to answer a given query. This

agent explicitly searches among the activities and re-

sources of the information source, manages their lo-

cation, and provide the required answers to the Router

Agent.

Agent interaction occurs under two main pro-

cesses.

Query, in which the Representative Agent re-

ceives the query from the user. The latter enriches the

query date with the user proﬁle (provided by the Rep-

resentative Agent) and the context data (provided by

the Context Agent). The Intermediary Agent receives

the enriched query, which categorizes the query ac-

cording to its type (SQL, keyword search, etc.). The

Router Agent receives the categorized query and ﬁnds

the relevant information sources and sends the query

to the corresponding Information Source Agent.

Results Retrieval, in which the Information

Source Agent executes the query and sends the re-

sults back to the Router Agent, then to the Intermedi-

ary Agent and to the Representative Agent. The latter

asks the Context Agent to determine if there were any

context changes. If there are no context changes, the

results are returned to the user, otherwise the query is

run again with the new context information.

4.2.2 Meta-Planning

PUMAS Lite selects a set of candidate activities for

the given course goals. The next step is create a meta-

plan, a compact structure that contains all of the ac-

tivities and is organized in such a way that all of the

adapted plans can be extracted from that structure.

In many cases, activity planning through algo-

rithms, such as GraphPlan, requires each activity to

have pre and post conditions, which are the conditions

required to perform the activity and the assertions that

must be true after successfully executing the activity,

respectively (Blum and Furst, 1997). In these kinds

of algorithms, the activity precedence is not known

before the planning algorithm is executed.

However, in the educational context, to deﬁne pre

and post conditions can be a tricky task. Teachers

must deﬁne several activities and ensure that they can

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

Figure 3: Meta-plan example.

be chained in many different ways to have a sufﬁ-

ciently broad set of different plans for the various stu-

dents in their courses. Moreover, in the educational

context pre and post conditions are often not strict. In

many situations course activities can be performed by

students even though they may have not fully meet the

activities’ requirements. In other cases, when course

activities require more strict pre-conditions, the stu-

dent may perform remedial activities before executing

the course activity

Taking all of the above into account, ASHYI-

EDU’s meta-plan uses a non-strict approach for pre

and post conditions and a slightly more structured

way to organize activities in the meta-plan. The es-

sential premise is that course goals are accomplished

in a speciﬁc sequence during the course. Therefore,

the precedence of course activities is given by the

goals to which they are aligned.

It is important to note that each activity may have

different alternative resources (e.g. ﬁles, URLs) that

the student may utilize in case the main resources are

not available. To represent this information in the

meta-plan, each node in the graph corresponds to a

combination of a given activity with one of the main

or alternative resource. In other words, the graph may

contain one or more nodes that correspond to the same

activity, but utilizing different resources.

To better illustrate the approach Figure 3 shows

an example of a meta-plan in ASHYI-EDU. Activ-

ities and resources are organized in a multi-partite

graph, in which each partition is denoted as a column.

Orange-colored nodes correspond to course activities,

while blue-colored ones correspond to remedial activ-

ities. A course activity partition is a column in the

graph containing orange-colored activities that focus

on the same course goal. A remedial activity partition

is a column of blue-colored activities that is placed in

between course activity partitions. The green node at

the left side of the graph is not an activity, but the

start node in the graph, the node from which all of the

adapted plans will begin.

Each partition is fully connected with the next par-

tition. In other words a course activity partition is

fully connected to the next remedial activity partition

and each remedial activity partition is fully connected

to the next course activity partition.

In addition, each course activity partition is also

fully connected to the next course activity partition.

In other words, a course activity partition that corre-

sponds to a given course goal is fully connected to the

course activity partition of the next goal. For simplic-

ity, these connections are not shown in the ﬁgure.

4.3 Creating Adapted Plans

The meta-plan holds all of the possible adapted plans

for the course. After creating the meta-plan, the next

step is to obtain an adapted plan for each student. An

essential piece of information in this stage is the stu-

dent proﬁle.

To obtain this information, ASHYI-EDU mea-

sures personality traits and learning styles of the stu-

dents. To measure personality, ASHYI-EDU uti-

lizes the Myers Briggs personality test (MentiScore-

Solutions-Limited, 2014), which proﬁles students ac-

cording to 16 personality traits. A student may

express these personality traits in different degrees.

To measure learning styles, ASHYI-EDU utilizes

the Honey-Alonso Learning Styles Questionnaire

(Alonso et al., 2009).

After the student answers the above tests, the stu-

dent proﬁle is complete and ASHYI-EDU is able to

create the adapted plan for the student.

Recall that an adapted plan corresponds to a path

from the start node on the left side of the meta-plan

graph to any node in the last column of the right (see

Section 4.2.2). Figure 4 depicts an adapted plan ob-

tained from the meta-plan and a student proﬁle. The

best plan for a given student corresponds to the short-

est path in the meta-plan graph, represented as the

non-gray nodes and the green connections. For illus-

trative purposes, the nodes in the optimal route are

zoomed in, but in the real application they have the

same size of the other nodes. The function that de-

termines the weights depends on student and activity

information. Therefore, edge weights will be differ-

ent from one student to another and the optimal path

might differ from one student to another. This ensures

that each student will obtain a personalized learning

plan.

The learning plan determines the order in which

activities should be performed by the student, starting

from the activities on the left side and ending on the

activities on the right side. Sometimes, activities may

require the student to previously develop some abili-

ties or competencies to properly execute those activ-

ities. ASHYI-EDU is able to detect when a student

ASHYI-EDU:ApplyingDynamicAdaptivePlanninginaVirtualLearningEnvironment

Figure 4: Adapted Plan Example.

does not meet those requirements, and plans accord-

ingly. In those cases, ASHYI-EDU will include re-

medial activities in the plan, so that the students can

develop those missing abilities and competencies be-

fore performing the activity that requires them.

Another important feature is that ASHYI-EDU

takes into account the changes in the student during

the execution of the plan. For instance, a student may

learn new abilities or competencies after successfully

executing a given activity. The adapted plan created

by ASHYI-EDU assumes that the student success-

fully executes every activity. Changes in his/her pro-

ﬁle after each activity are taken into account to deter-

mine the afﬁnity with the remaining activities in the

meta-plan.

If, for some reason, the student does not success-

fully executes an activity, ASHYI-EDU is able to cre-

ate a new plan that takes into account this situation

(see Section 4.6).

4.4 Matching Students to Activities

To determine the best learning plan for a given stu-

dent, ASHYI-EDU must select the most suitable ac-

tivity for the student to accomplish each learning goal.

This selection is based on the afﬁnity between the stu-

dent and each activity. This afﬁnity if obtained from

the edge weights in the meta-plan graph. The weight

of an edge between two nodes in the graph represents

the afﬁnity between the student and the activity to

which the edge is pointing (i.e., the activity on the

right side of that edge).

To calculate the edge weights, ASHYI-EDU uti-

lizes a distance function that yields a value that is in-

versely proportional to the degree of afﬁnity between

a student and an activity. Consequently, the shortest

path in the meta-plan will yield the set of activities

that have the most afﬁnity to a given student.

The distance function takes as input a student vec-

tor and an activity preconditions vector. Both vec-

tors have the same structure, summarized in Figure

5. Each vector has 55 components, each one repre-

senting a speciﬁc attribute. The values range from 0,

Figure 5: Student and Activity Vector.

which means the total absence of a given attribute, to

1, which means a strong presence of the attribute. The

ﬁrst 4 components correspond to learning styles, the

next 16 components are personality traits, the next 19

are abilities, and the last 16 are competencies.

For students, the vector components represent the

degree of a certain learning style, personality trait,

ability, or competency is present in each student. For

instance, a value close to 1 in a learning style means

that the student has a strong preference to learn using

that style. Similarly, a value close to 1 in a person-

ality trait means that the student strongly manifests

that trait. A value close to 1 in an ability or compe-

tency means that the student has signiﬁcantly devel-

oped that activity or competency.

In an activity, the vector represents a non-strong

set of preconditions. To be more precise, the activity

vector represents the degree of suitability of the ac-

tivity for students who have a certain set of attributes.

For instance, a value of 1 in a learning style or per-

sonality trait means that the activity is highly suitable

for students who have a high preference for that learn-

ing style or strongly manifest that personality trait. A

high value in an ability or competency means that it is

highly recommendable that the student develops that

ability or competency to successfully execute the ac-

tivity.

Some vector components have continuous values

between 0 and 1 (learning styles), while others may

assume only the values 0 or 1, but no values in be-

tween (personality traits, abilities, and competencies).

Because of these differences, the distance function d

is a weighed average of different distance functions:

d(S, A) = w

(S, A) +w

(S, A) +

(S, A) +w

(S, A)

Where S is the student vector, A is the activity vec-

tor, d

is the euclidean distance between learning style

components of S and A, d

is the euclidean distance

between personality trait components of S and A, d

the Jaccard (Candillier et al., 2008) distance between

ability components of S and A, d

is the Jaccard dis-

tance between competency components of S and A,

and each w

corresponds to the weight of each dis-

tance metric in the ﬁnal result.

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

Figure 6: Adapted Plan as a List.

4.5 Adapted Plan Execution

Figure 4 is an example of the way ASHYI-EDU dis-

plays an adapted plan for a student. The adapted

plan corresponds to the optimal route in the meta-plan

graph and represents the set of activities that are more

alike to a student.

The student can also visualize his/her own adapted

plan as a list, as shown in Figure 6 (the ﬁgure is

cropped for space reasons). Different icons denote the

various states of each task. Pending activities have an

hourglass icon. Available activities are marked with a

pen icon. Activities that have been ﬁnished have ei-

ther a happy face or a sad face icon, depending on the

teacher’s feedback.

The student can execute this plan following these

steps:

1. Access the activity form in Figure 7 and download

the resources to execute the activity, which can be

ﬁles or URLs.

2. Follow the instructions and execute the activity.

3. Access the activity form (see Figure 7) and com-

plete two ﬁelds: the ﬁle to upload with the an-

swer to the activity and the feedback to send to

the teacher.

4. The teacher evaluates the activity thereafter and

provides feedback to the student through the same

form. The teacher can indicate the skills and com-

petences achieved by the student, the grade, and

the feedback to the student.

5. The student then receives the teacher’s feedback

and can execute the activities that become avail-

able after the teacher’s evaluation.

6. The above steps are repeated until the student has

executed the entire adapted plan.

4.6 Context and User Changes and

Re-Planning

In some cases, things may not occur according to

plan. For instance, the student may not successfully

execute an activity, which means that he/she does not

Figure 7: Activity Form.

learn the expected abilities or competencies. In other

cases, context changes may limit the access to certain

resources. For instance, the student may be trying to

access the activity resources from outside the univer-

sity and may not have access to services protected be-

hind the university’s ﬁrewall. These situations may

require to provide alternative activities or resources

so that the student can execute the plan successfully.

To address the above issues, ASHYI-EDU is able

to automatically detect changes in the student proﬁle

or in the context and create a new plan. To perform re-

planning, ASHYI-EDU updates the user proﬁle and

the meta-plan with the corresponding changes and ex-

ecutes the planning algorithm with this new data. The

result is a plan that is best adapted to the new scenario.

Currently, ASHYI-EDU is able to perform re-

planning based on the following scenarios:

1. The student did not learn the expected abilities or

competencies after the execution of a given activ-

ity. In this case, ASHYI-EDU updates the student

proﬁle accordingly and creates a new plan that in-

cludes activities more alike to the student in this

new scenario. This new plan may also include ad-

ditional remedial activities.

2. The student is not able to access a given resource.

For instance, if the student is outside the univer-

sity and he/she needs to access an internal service

of the university that is protected by a ﬁrewall

from the outside. To address this case, ASHYI-

EDU disables the nodes that contain the inacces-

sible resources and executes the planning algo-

ASHYI-EDU:ApplyingDynamicAdaptivePlanninginaVirtualLearningEnvironment

rithm again. Recall that multiple nodes in the

meta-plan graph may contain the same activity,

but with different resources (see Section 4.2.2).

Therefore, some activities can still be included in

an adapted plan if they are associated with avail-

able resources.

3. The teacher decides to cancel a speciﬁc activ-

ity or make it unavailable to students. In this

case, ASHYI-EDU removes all of the correspond-

ing nodes from the meta-plan graph and executes

the planning algorithm again. Recall that sev-

eral nodes in the meta-plan graph may have the

same activity, but different resources (see Section

4.2.2). Therefore, all of the nodes containing that

activity must be removed from the graph.

5 CASE STUDY

To validate ASHYI-EDU, we are currently develop-

ing a case study in a university course for students of

the career of Primary School Teacher. The course is

called "Learning to Learn in the Web". This course

utilizes blended learning and its goal is to give the

students tools to learn in the web.

This course is being taught each semester to differ-

ent groups of students.The case study spans through

three semesters. The ﬁrst semester (Spring 2014)

included an ofﬂine course that did not use ASHYI-

EDU and relied only on manual procedures to create

adapted plans and interact with students. The sec-

ond semester (Fall 2014) includes an online course

that utilizes ASHYI-EDU to manage the entire course

planning and student-teacher interaction. The third

semester (Spring 2015) will include a course utilizing

ASHYI-EDU in the same way as the second semester.

5.1 Course Structure

Figure 8 denotes the structure of the course in all of

the three semesters. The course has several goals that

students should satisfy after ﬁnishing the course. To

satisfy those goals, the course comprises three learn-

ing units. Each learning unit has a set of goals that

must be satisﬁed and a meta-plan with the activities to

satisfy those goals. The ﬁrst learning unit has 4 goals,

the second has 3 and the third one has 4. Each activ-

ity in the meta-plan is either a course activity aimed to

satisfy one of those objectives or a remedial activity.

5.2 Ofﬂine Course

This course is called ofﬂine because it was taught

face-to-face to students without the help of a virtual

Course

Learning Unit 1

Goal 1

Learning Unit N

Goal M

...

Activity 1 Activity n

...

Figure 8: Course Structure.

Table 2: Student Characteristics in the Ofﬂine Course.

Stu-

dent

Ac-

tive

Prag-

matic

Reﬂe-

xive

Theor-

etical

Perso-

nality

E1 7 8 12 8 ENTJ

E2 10 17 17 18 ENFJ

E3 17 16 18 17 ESFJ

E4 7 18 19 16 ENFJ

E5 12 12 10 6 ENFJ

E6 7 14 18 14 ENFJ

E7 14 15 16 17 ENTJ

E8 10 13 13 13 ENTJ

learning environment. The teachers evaluated stu-

dents and created adapted plans manually.

The course had 32 students, from which only 8

ﬁlled the tests to deﬁne their proﬁles. They were from

different careers: Middle School Spanish Teaching

Licentiate, Accounting, Psicology, and Information

Sciences.

Table 2 shows the characteristics of the students

who took that course. Columns 2 to 5 correspond to

learning styles scores, where the maximum value is

20. The last column correspond to personality traits,

where E means Extroverted, N means Intuitive, S

means Sensing, T means Thinker, F means Feeling,

and J means Judging.

As shown in that table, there are students who

have different learning styles and personality traits.

Most students have high preferences for Reﬂexive and

Theoretical styles, while fewer ones prefer Pragmatic

and Active styles.

Table 3 summarizes the results of the execution of

the ﬁrst learning unit. Grades are in a scale of 0 to 5.

The teachers determined that the low performance of

some of them was because of lack of reading compre-

hension and writing skills. Therefore, they proposed

a set of remedial activities to address these issues.

5.3 Online Course

The online course was performed the following

semester after the ofﬂine course (Fall 2014). Most

of the online course is based on the ofﬂine course.

The online version adds more activities, so that each

learning goal can have several alternative activities to

achieve it.

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

Table 3: Results of the ﬁrst learning unit.

Student Grade Grade Analysis

E1 3 Did not execute all activities

E2 3.2 Incomplete work. The student didn’t

do what was asked for

E3 1 Incomplete work. The student didn’t

do what was asked for, writing errores

E4 5

E5 5

E6 4.6 Does not clearly express ideas

E7 4.4 Does not clearly express ideas

E8 2.5 Incomplete work. The student didn’t

do what was asked for. Does not

clearly express ideas

The study of that semester involved all of the

course 29 participants: 6 men and 23 women. The

students were from different backgrounds: Education,

Accounting, Information Sciences, Psychology, Soft-

ware Engineering, Business Administration, and Mi-

crobiology.

A prototype of ASHYI-EDU was created, which

utilizes Sakai (Apereo-Foundation, 2014), a Java-

based virtual learning environment software. The im-

plementation of ASHYI-EDU relies on a modiﬁed

version of Sakai’s Lesson Builder component that in-

cludes dynamic activity planning to build learning

units and its lessons and a simple interaction facility

for students and teachers.

The development of the online course focuses on

automatically performing several tasks that were slow

or cumbersome in the ofﬂine course: i) providing

adapted plans according to the necessities of each stu-

dent, ii) automatically re-planning according to con-

text changes, and iii) register the interaction between

users and the system.

The ASHYI-EDU prototype provides several fea-

tures to facilitate the creation of dynamic adaptive

plans and monitor their execution. Recall Figure 3

that shows the meta-plan of a learning unit that is

built by the teacher. The teacher can create, modify or

remove activities from the meta-plan and the system

will automatically organize them according to learn-

ing goals and activity type (course or remedial).

A student that begins a course must complete the

Myers Briggs personality test and also the Honey-

Alonso Learning Styles Questionnaire tests. ASHYI-

EDU uses this information to ﬁll the student proﬁles

and to automatically create adapted plans for each stu-

dent.

ASHYI-EDU is able to provide different plans for

different students, according to their speciﬁc charac-

teristics.

Figure 9: Adapted Plan for Student A.

Figures 9 and 10 correspond to adapted plans for

Student A and B, respectively. For space reasons the

chosen examples are not radically dissimilar but in

practice two students may have very different plans.

In this example they have two important differences:

Remedial activities: ASHYI-EDU did not plan

remedial activities for the Student A, while it planned

a remedial activity, called Information Search and Se-

lection (shown in blue) for Student B. This activity

aims to develop analysis and synthesis abilities in the

student. ASHYI-EDU assigned the remedial activity,

because the student lacked these abilities for the ﬁrst

course activity.

Course activities: To accomplish the ﬁrst learn-

ing goal (second column in both graphs), to analyze,

reﬂect and understand the information search pro-

cess, ASHYI-EDU assigned Student A a paper read-

ing activity to develop reading comprehension abili-

ties, while the Student B must perform information

search to learn critical thinking and analysis and syn-

thesis abilities. These activities were assigned accord-

ing to the student learning styles and personalities:

Student A: Learning Style: Active - level:

6, Pragmatic - level: 7, Reﬂexive - level: 15,

Theoretical- level: 12; Personality: INFJ; Skills: un-

derstanding, relate their reality with the environment,

interpret and analyze information, observation, gen-

erates own answers from their knowledge and experi-

ence.

Student B: Learning Style: Active - level:

15, Pragmatic - level: 19, Reﬂexive - level: 19,

Theoretical- level: 19; Personality: ENFJ; Skills:

understanding, observation, generates own answers

from their knowledge and experience, shows interest

and initiative to continue learning, agility and adapt-

ability, empathy, and global vision, interpersonal re-

lationships and managing emotions and feelings.

ASHYI-EDU:ApplyingDynamicAdaptivePlanninginaVirtualLearningEnvironment

Figure 10: Adapted Plan for Student B.

5.4 Current Results

Students were interviewed to determine their satisfac-

tion with the online course. The overall responses

were positive. The students expressed that the course

helped them to understand the way they approach to

learning in the course, which learning styles are the

most important in their learning process, and the most

adequate activities for them. The students felt more

identiﬁed and comfortable with the activities that they

performed. Similarly, they felt that the inclusion of

remedial activities helped them to improve their per-

formance.

The online course has only been applied to one

group of students. The case study will yield more de-

tailed results after ﬁnishing the Spring 2015 course,

which will yield results that could be compared with

Fall 2014.

From the point of view of the teachers, one impor-

tant change is the amount of effort to create the course

content. While a traditional course may need only

one type of activity per learning goal, ASHYI-EDU,

requires several activities per learning goal. Another

important change is that teachers require to charac-

terize each activity to determine how alike are them

to speciﬁc types of students. The above requires a

high initial effort to create the course. However, the

repository provided by ASHYI-EDU has the potential

of reducing this effort by facilitating reuse of existing

material.

6 CONCLUSIONS AND FUTURE

WORK

This paper presented ASHYI-EDU, a system to per-

form dynamic adaptive planning in virtual learning

environments. ASHYI-EDU captures the essential

student characteristics through a series of tests and is

able to create adapted plans that are the most aligned

with those characteristics. ASHYI-EDU is also aware

of changes in context and is able to dynamically mod-

ify existing plans to make better plans for new con-

straints.

The feedback obtained from the students suggest

that ASHYI-EDU is effectively assigning the most

alike activities to each student, based on their speciﬁc

abilities, competencies, personality traits, and learn-

ing styles. The feedback of the students about reme-

dial activities suggests that the automatic assignment

of remedial activities effectively addresses learning

needs of the students. Overall, it means that ASHYI-

EDU has the potential of improving the teaching-

learning process.

It is important to note that the student-activity

matching process is as good as the information in the

student and activity proﬁles. Further validation of the

proposed approach requires: to ensure that the tests

utilized to build the student proﬁle are the most ef-

fective ones to capture the student characteristics, and

that the activity proﬁles are effectively the most ade-

quate for certain types of students.

Overall, the proposed system is able to provide

a ﬁne-grained personalized learning plan to hetero-

geneous students, which takes into account more in-

formation than existing approaches. Another advan-

tage over related work is the ability to re-plan based

on context and user information, while ensuring that

each student has the best learning plan possible at all

times. Lastly, the ability to select remedial activities

for students facilitate even more the learning process,

to ensure that student have the required abilities and

competences to complete all activities in a learning

unit.

Future work is to complete the ﬁrst and second

semesters of the online course, analyze their results

and ﬁnish the validation of the proposed approach.

ACKNOWLEDGMENTS

This paper is part of the project “ASHYI: Plataforma

basada en agentes para la planiﬁcación dinámica, in-

teligente y adaptativa de actividades aplicada a la ed-

ucación personalizada”, executed by the ISTAR re-

search group of the Pontiﬁcia Universidad Javeriana,

coﬁnanced by Colciencias, project #1203-569-33545.

REFERENCES

Alonso, C., Gallego, D., and Garcia, J.

(2009). CHAEA - estilos de aprendizaje.

http://www.estilosdeaprendizaje.es. Last accessed:

2013-10-17.

AnousouyaDevi, M., Uma, V., and Aghila, G. (2012). Tem-

poral planning with reference event based temporal re-

lations. In 2012 International Conference on Recent

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

Trends In Information Technology (ICRTIT), pages

548–552.

Apereo-Foundation (2014). Sakai. https://sakaiproject.org/.

Last accessed: 2013-09-24.

Bahmani, A., Sedigh, S., and Hurson, A. (2011). Context-

aware recommendation algorithms for the percepolis

personalized education platform. In Frontiers in Edu-

cation Conference (FIE), 2011, pages F4E–1–F4E–6.

Blum, A. L. and Furst, M. L. (1997). Fast planning through

planning graph analysis. Artiﬁcial Intelligence, 90(1-

2):281–300.

Candillier, L., Meyer, F., and Fessant, F. (2008). Design-

ing speciﬁc weighted similarity measures to improve

collaborative ﬁltering systems. In Perner, P., editor,

Advances in Data Mining. Medical Applications, E-

Commerce, Marketing, and Theoretical Aspects, num-

ber 5077 in Lecture Notes in Computer Science, pages

242–255. Springer Berlin Heidelberg.

Carrillo-Ramos, A., Villanova-Oliver, M., Gensel, J., and

Martin, H. (2007). Knowledge management for

adapted information retrieval in ubiquitous environ-

ments. In Filipe, J., Cordeiro, J., and Pedrosa, V.,

editors, Web Information Systems and Technologies,

number 1 in Lecture Notes in Business Information

Processing, pages 84–96. Springer Berlin Heidelberg.

Coles, A., Coles, A., Fox, M., and Long, D. (2010).

Forward-Chaining Partial-Order Planning.

Fikes, R. E. and Nilsson, N. J. (1972). STRIPS: A new ap-

proach to the application of theorem proving to prob-

lem solving. Artiﬁcial intelligence, 2(3):189–208.

Geib, C. W. and Goldman, R. P. (2009). A probabilistic plan

recognition algorithm based on plan tree grammars.

Artiﬁcial Intelligence, 173(11):1101–1132.

Guoqi, X., Jun, P., Xiaoyong, Z., and Kuo-Chi, L. (2010).

A dynamic task planning based on task subcontract-

ing and dynamic re-decomposition. In Control Con-

ference (CCC), 2010 29th Chinese, pages 4518–4523.

Jaime Pavlich-Mariscal, Yolima Uribe, L. F. B.-L. N. A. M.-

M. A. C.-R. A. P. Q. R. M. V. R. F.-G. S. B. (2015). An

Architecture for Dynamic and Adaptive User Activity

Planning Systems. In Webist.

Johnson, P. A. and Sieber, R. (2009). Agent-based mod-

elling: A dynamic scenario planning approach to

tourism PSS. In Geertman, D. S. and Stillwell, P. J.,

editors, Planning Support Systems Best Practice and

New Methods, number 95 in The GeoJournal Library,

pages 211–226. Springer Netherlands.

Li, S., Li, G., and Chen, Y. (2008a). A study on informa-

tion push in personalized education system. In Inter-

national Symposium on Knowledge Acquisition and

Modeling, 2008. KAM ’08, pages 534–537.

Li, S., Li, G., and Chen, Y. (2008b). A study on informa-

tion push in personalized education system. In Inter-

national Symposium on Knowledge Acquisition and

Modeling, 2008. KAM ’08, pages 534–537.

Marki, F., Charypar, D., and Axhausen, K. W. (2012). Tar-

get driven activity planning.

MentiScore-Solutions-Limited (2014). 16 personali-

ties. http://www.16personalities.com. Last accessed:

2014-10-06.

Mohamad, F., Mumtazimah, M., and Fadzli, S. (2013).

Integrating an e-learning model using IRT, felder-

silverman and neural network approach. In 2013 Sec-

ond International Conference on Informatics and Ap-

plications (ICIA), pages 207–211.

Nijland, L., Arentze, T., and Timmermans, H. (2012). In-

corporating planned activities and events in a dynamic

multi-day activity agenda generator. Transportation,

39(4):791–806.

Oh, H., Kim, H.-C., and Hong, K.-W. (2009). A dynamic

perspective of meeting planners’ satisfaction: Toward

conceptualization of critical relevancy. Tourism Man-

agement, 30(4):471–482.

Plaku, E. and Hager, G. (2010). Sampling-based motion

and symbolic action planning with geometric and dif-

ferential constraints. In 2010 IEEE International Con-

ference on Robotics and Automation (ICRA), pages

5002–5008.

Rytikova, I. and Boicu, M. (2014a). A methodology for per-

sonalized competency-based learning in undergradu-

ate courses. In Proceedings of the 15th Annual Con-

ference on Information Technology Education, SIG-

ITE ’14, pages 81–86, New York, NY, USA. ACM.

Rytikova, I. and Boicu, M. (2014b). A methodology for per-

sonalized competency-based learning in undergradu-

ate courses. In Proceedings of the 15th Annual Con-

ference on Information Technology Education, SIG-

ITE ’14, pages 81–86, New York, NY, USA. ACM.

Sangineto, E., Capuano, N., Gaeta, M., and Micarelli, A.

(2008). Adaptive course generation through learning

styles representation. Universal Access in the Infor-

mation Society, 7(1-2):1–23.

Yang, Q. and Chen, L. (2013). A learning grouping algo-

rithm based on user personality. In 2013 8th Inter-

national Conference on Computer Science Education

(ICCSE), pages 71–75.

ASHYI-EDU:ApplyingDynamicAdaptivePlanninginaVirtualLearningEnvironment