Internet growth has induced the development of a large number of collaborative tools for online writing and information sharing. Educators quickly realized the bene ts of such tools for learners, allowing them to work online, to share their knowledge and help each other. Distance learning is a key concept in today's educational research; collaborative learning environments are becoming widespread, being more dynamic and resourceful. However, distance learning also introduced a series of problems, such as high evasion rates resulting from lack of support and personalized feedback. It has also introduced di culties for educators to follow and reviews students' assignments. Based on this scenario, the work presented here proposes the development of a pedagogical agent supported by an intelligent tutoring system to provide students and teachers with assistance in order to minimize some of these problems. The use of a pedagogical agent allows students to have a constant feedback and guidance based on the identi cation of problems that may emerge from an online collaborative writing activity. The presence of this agent is intended to help students coordinate their e orts in writing a text collaboratively, and improve their work in terms of coherence. Furthermore, the pedagogical agent is also be able to assist teachers, reporting problems and simplifying their tasks related to the analysis of the work produced by each student. To support our pedagogical agent we propose the use of data mining tools to extract information related to the students' writings, and a recommender system to suggest additional resources.
Alongside with the internet growth and online collaborative
learning environments, data mining has become more
important and popular in the education eld. From the student's
perspective, the possibility to easily search for learning
material using indexes and other reference tools increases their
resources at the same time that reduces the e ort needed to
nd relevant information. From the teachers' perspective, it
can help them to summarize students' writings in a distant
learning context [
The possibility to work collaboratively is attractive both
for students and teachers. It allows students to exchange
knowledge, to help each other and complement their work
with di erent ideas. However, it also creates several di
culties for teachers when evaluating the work done, once it's
hard and demanding to monitor each student production
[
This work has also been motivated by the fast increase in the number of collaborative writing tools available online, and the problems that often originate from their use. For instance, the lack of interaction and actual collaboration in collaborative writing tasks, and the lack of supervision and feedback to students' work.
This project proposes a pedagogical agent to be used in a collaborative writing environment, so that it may assist teachers and students in their tasks. The agent uses data mining to identify problems in the students' writings, and based on this information it tries to guide them in improving their collaborative text production. Besides, the agent gives teachers more accurate information about the students' participation, their di culties and interactive writing process.
Intelligent tutoring systems (ITS ) are computer tools
capable of providing customized instruction or feedback to
learners [
ITS often uses a variety of computational resources for
analyzing the users interaction with the system. These are
adaptive mechanisms, capable of personalizing learning
according to individual student characteristics, such as
knowledge on the subject, mood and emotion [
However, the ability to properly help the user often depends
on the interface between the system and the user. Not rarely,
ITS require a virtual character to interact with users. These
characters are called pedagogical agents. The use of
pedagogical agent (PA) in educational applications has
demonstrated that these animated characters may improve
student's engagement and learning experience [
Among the many bene ts of using PAs are the increase of
motivation, perception of ease and comfort in the
learning environment, the promotion of fundamental behaviors of
learning, the realization of a need for personal relationships
in learning and gains in terms of memory, understanding
and problem solving [
Based on the highlighted di culties of the development of collaborative work online, we propose a pedagogical agent to be inserted in an online collaborative writing environment. This agent will be capable of helping students through immediate feedback about their collaborative text production, and it will assist teachers through the presentation of information/indicator regarding students' participation and progress in the assigned tasks. Our goal is to provide a full time assistance to the users of the collaborative environment and to reduce the amount of work needed to analyze their interactions and work.
In order to do that, we designed a pedagogical agent to be integrated in the intelligent tutoring system. The ITS will be responsible for collecting all information regarding the students activities in the environment. The student could be adding text, images, audio, video or any other resource to the project or simply reviewing and modifying some previous work. In any case, the ITS must keep a log of those interactions in order to determine which action to perform (when needed). Among the di erent information collected by the system, we may list: the student's contribution to the project (either by the addition of new contents or revision/edition of previous work); frequency (how many times each student accessed the collaborative environment, for how long and when) and a concept map summarizing what has been written by each student. Once we are considering a collaborative environment, several users may be at the system at the same time and it is important for the ITS to identify which student is responsible for which action. Whether the students access the environment simultaneously or separated, it is important for all users to have an identi er that will inform the system which user is current online and modifying any given document.
All the information collected by the ITS will be processed
using data mining techniques. This allows the system to
identify what type of contribution the user has made to the
project and infer how cohesive and coherent the text
produced collaboratively is. A data miner similar to Sobek[
There are several types of interventions planned for both students and teacher. The most common type of interventions for students is a direct message sent by the PA inquiring about some aspect of the project. Those inquiries are intended to foster critical thinking and help students correct what the ITS identi es as a problem. The messages are sent when the students' work is incomplete or lack cohesion. In both cases, it is possible for the agent to suggest additional material that may help them correct the problem. Another possible intervention is the use of e-mail messages when students are not participating in the collaborative work, or when their contributions are not coherent with the remaining project. The last type of intervention is not an automatic answer from the PA, but an explicit request for help from a student. Speci c functions are being developed to enable students to ask the PA for further information about some aspect of the project, or about its structure and coherence.
A key feature in those interactions between PA and students
is the agent's ability to identify additional resources that
may help the students' text production. As the students'
project may include several types of media and di erent
format of resources, the agent is able to recommend students
with additional learning objects extracted from repositories
as well as from the web. Learning objects are usually
indexed using metadata, where keyword and object
descriptions are very common, information that is used to search
for learning objects that are most related to the topic at
hand. Using a similar technique to the one used by Breno
et. al. [
Some of the PA's functions are speci c to help teachers. As it is often di cult and time consuming for teachers to analyze the individual production of each student in collaborative writing tasks, the PA will provide accurate information about each student's contributions and progress based on the information collected by the ITS. This will allow teachers to identify relevant contributions made by each student or the absence of a student in a particular subject or part of the project.
The interaction between the PA and the teachers will be
set in two ways. The rst one is through direct messages
sent by the PA to the teacher's email or other
communication method. Those messages are sent to inform about lack
of participation from some student or if the system detects
that one or more students need further assistance with the
project (this may be identi ed through a constant request for
help from a student, when his/her work is always identi ed
to have coherence/cohesion problems, or when the student's
work is constantly edited by other students). The second
form of interaction will be a support interface, where the
teacher will be able to request speci c information
regarding student's activities in the environment. All those PA
features are meant to provide a more personalized contact
between the agent, the students and/or teachers.
Another aspect the agent is concern about is the coherence
of the project. The students may not be together when
writing the project and it may result in disjoint texts,
unrelated or redundant information. Therefore, it is important
for the agent to identify coherence problems and contact the
students who produced the incoherent parts of the project.
The agent may also use the information gathered from those
incoherent parts to search for additional material and
learning objects that could help students ll the possible gaps in
the project. Although we considered using Latent
Semantic Analysis to perform this task [
In the current state of the project, we are developing a script program that will collect the data from students' writings and send it to the pedagogical agent. This is part of the development of our intelligent tutoring system. This is also one of the most challenging parts of this project, as it is important to correctly interpret and evaluate multimedia resources. Through the use of scripts without a particular user interface, we intend to create a more reusable system, allowing it to be used in many tools and environments without the need for changing the program's core.
Our data mining system based on Sobek is already being modi ed to provide useful information regarding resource similarity, concepts and relationships. The input for our data miner is very restrictive, but we are working on making it more general. This is most useful to investigate how to successfully mine multimedia resources. Using the learning object repository, we may conduct experiments that will assess the quality of the results and the reliability of our data mining tool.
The experiments will be carried out in projects using Google Drive 1. The choice for this environment has been the extensive database and number of projects that are developed with this tecnology daily. It is one of the most known and complete environment for collaborative writing, and it also allows the input of scripts that could facilitate the integration of our pedagogical agent with the system. Google support and APIs make it a natural choice for our project to be integrated in Google Drive Documents, to ensure the dissemination of our ideas and software to a larger number of people across the world. 1https://drive.google.com Daniel Epstein has Master Degree in Computer Science from Universidade Federal do Rio Grande do Sul working with robotics and arti cial intelligence. He is a Ph.D. student at the Post Graduation Program in Computer Science in Education, currently researching Data Mining and Pedagogical Agent use on collaborative learning. He started the program in March 2013 and is currently writing his thesis proposal. He will complete his PhD in the middle of 2016. Eliseo Berni Reategui has a PhD degree in Computer Science from the University of London, England, and MS in Computer Science from Universidade Federal do Rio Grande do Sul, Brazil. After nishing his PhD and working in the industry for 5 years, Dr. Reategui held a lecturer position at the University of Caxias do Sul, Brazil, for a few more years. Nowadays, he works as a lecturer and researcher at the Federal University of Rio Grande do Sul. His research interests are related to the use of computers in education, involving areas such as arti cial intelligence and human-computer interaction. The main topics on which the student would like to receive advices are: