Accessing OER from MOODLE through X5GON’s AI tools
Authors: Walid Ben Romdhane (Université de Nantes, France), Colin de la Higuera, (Université de Nantes, France), Davor Orlic (IRCAI, Slovenia), Kristijan Perčič (Posta Slovenije, Slovenia)
Corresponding author: Colin de la Higuera, cdlh@univ-nantes.fr. Laboratoire des Sciences du Numérique de Nantes, Université de Nantes. France.
Abstract :
Project X5GON is a European project (2017-2020) in which are being developed tools allowing to access, transcribe, translate, index, search for and recommend OER from multiple languages, topics, repositories. In order to better use these features a Moodle plug-in has been produced. When this plug-in is installed, a teacher can benefit from an environment in which he can give access to his classroom to a search engine which will adapt to the classroom’s preferences, recommend new resources based on the classroom’s positive feedback and a playlist designed by the teacher with his personal choice of OER.
1. Context
Open Educational Resources (OER) allow teachers to reuse, revise, remix, redistribute courseware produced by others. Their adoption is progressively stepping forward, with an important landmark in the adoption, in November 2019, by the General Assembly of Unesco, of its recommendation. If political, financial, legal difficulties still exist for their adoption, there are also less visible difficulties linked with technical issues. Artificial intelligence may contribute to lift some of these technical barriers. This has been the goal of the teams involved in project X5GON.
X5GON is a state of the art Artificial Intelligence powered recommendation engine for OER materials. It uses a portfolio of human-centered artificial intelligence methods, algorithms and tools which are unobtrusive and away from the eye of the users, thus making it intuitive and easy to use, as well as making very clear the different usages of data, ensuring the necessary transparency. It works on two levels, first it crunches educational materials data, estimates the quality of these materials and difficulty levels and if connected into a network of OER sites, it calculates and recommends a material to the most appropriate learners in a way that it will benefit their learning journeys. X5GON has already delivered a portfolio of 20 prototypes and applications, including a mobile app and a number of APIs that allow software interoperability. Currently it connects 117,781 materials to 2.2M users and has facilitated 10.7 user-material learning interactions., 13 repositories of which it has data, and 4 repositories of which it has user visits data, counting a total of 370,069 material contents, these are automatically transcribed and translated with native AI so that effectively more than half of these are material translations into other languages, including Slovenian, English, Portuguese, Spanish, Catalan, French, and German. This technology is specifically designed to help UNESCO member states implement in the easiest possible way the new UNESCO Recommendation on OER.
Moodle is the most important Learning Management System in the world today. It is believed to be used by 80% of the Universities, but it also is used in schools. The total number of sites over which it is deployed exceeds 150,000 over 242 countries. Teachers use Moodle to interact with their students, by giving them access to resources, allowing interactive activities, extracting learning analytics and using it for grading purposes.
Artificial Intelligence is sometimes described as the new electricity. It is being applied to many fields, including education. Automatically extracting meaning from texts, which themselves will effortlessly have been obtained from videos, audios and documents in a variety of formats is today possible. Other applications of AI for education include intelligent tutors, learning analytics, automatic grading…
Unesco is an important worldwide actor when it comes to education. It has promoted open education resources and recently adopted a recommendation which -it is hoped- will further increase the offer of new courses and the demand for these. In turn, this emphasizes the importance of having the right tools to help the ecosystem to function. Project X5GON have been working closely with UNESCO: three of the teams from the project hold UNESCO chairs on AI and/or OER.
2. The AI inside X5GON
In a nutshell, Open Educational Resource Repositories are found over the web, identified as such (the license is checked) and crawled. The ingestion process passes the OER to powerful transcription and translation tools [4]. The result is raw text: the current state of technology allows this text to be surprisingly clean, with an acceptable number of errors. This text can then be indexed. And a number of features are extracted through the use of models built by Machine Learning algorithms: Keywords and topics are obtained, the difficulty of the text is evaluated, some form of precedence is computed. These semantic features can now be used for more complex and rich tasks.
Search engines can just work on patterns (the presence or absence of keywords), but they can also work on more complex similarities. Recommendation engines take the knowledge the system has of the current resource being consulted and, through a mechanism called collaborative filtering, suggests alternative documents which could be of use, provided the user is looking for the topics identified so far. Alternatively, if we know enough about the user, the system can recommend resources close to those he has shown interest in. Obviously, this does mean that the user has volunteered this information in some way.
When dealing with online education, system developers are facing a conundrum: on one hand they want to offer the best service and this involves personalization: each teacher, each learner is different, and adapting recommendations to suit the identified needs of the teacher/student is a legitimate concern. On the other hand the traces left by learners can be very meaningful and could be used in several non desired situations. Furthermore, these traces cannot (and should not) be gathered that easily: in Europe, important decisions concerning this (notably the General Data Protection Regulation -GDPR- dispositions) regulate strongly, and other countries in the world also have built laws protecting the individual users.
But Artificial Intelligence, and more notably Machine Learning, strives on user data. So how can we get the better of both worlds: respect the user’s privacy and provide the machine learning algorithms with the data it needs to build the models which will ultimately propose the best solutions to the users?
3. The Moodle tool
Many of the tools developed in project X5GON cannot be included as stand-alones inside Moodle: there are technical constraints related with the size of the models and the compatibilities between security restrictions.
But we have developed a plug-in which can be installed by any Moodle administrator and does give access to three of X5GON’s main features.
3.1 The Moodle tool from the point of view of the teacher
A teacher can access the course she is teaching and add the new X5GON activity (see Fig. 1). This activity comes in three variants: X5-Discovery, X5-Recommend and X5-Playlist. In order to seed the system the teacher will use X5-Discovery herself and enter some keywords she would use to find new material related to the course or the fragment of the course the teacher wants her students to work on. The teacher can choose which of the three X5 tools she shares with her students. If the teacher chooses X5-Playlist and has not built her playlist, she is invited to go to the x5learn.org page to do so.
Fig. 1 The interface allowing a teacher to choose the X5GON activity
3.2 The Moodle tool from the point of view of the student
Once the X5-GON activity is installed, a student from that class can search for resources using one of three settings.
With X5-Discovery, the student will search for new OER. He will be able to see the most popular searches made by his fellow students; this will give him a starting point. These most popular searches for the course are of course updated with his own searches.
Fig. 2 X5-Discovery with the OER suggested and also the keyword trends
With X5-Recommend, the student will get recommendations of OER based on his current work. This recommendation is built as a linear combination of the recommendations made by the X5GON tool and the more popular OER viewed by the classroom. Again, the recommendation is systematically updated as the student explores but also as the classroom shifts its attention to new OER.
Fig. 3 The recommendations
With X5-Playlist the student can access the different OER previously selected by the teacher. The number of accesses is registered giving teachers an important information concerning the extra material viewed by her students.
Fig. 4 The teacher’s playlist, with the added information
3.3 Some learning scenarios
A teacher may just want her students to find extra material. She may want the students to help each other and explore material that their fellow students will find. Or she may want to grade students who have found original material.
A teacher may want to propose to her students a specific set of educative resources as an initial support materials to start a brainstorming activity or discussing about the main topics detailed in the materials. Additional information is provided with the playlist items to give some insights for students about the material contents such as the concepts, keywords…
A teacher may also want to prepare the next iteration of her lecture by encouraging students to use the X5GON tools: the students will find new material which is automatically curated -at least by checking how the students have used it- and the teacher can incorporate this material next year.
3.4 Two key ideas
The first key idea is that students are not always that good at using internet navigating and search tools. They get easily lost, swamped by the information, incapable of deciding if a new document is of value. Through the AI, documents accessed and consumed by the rest of the class are recommended in priority.
The second key idea is that instead of picking user data, the system will collect usage data. At no moment will the system -outside Moodle- know which individual student performed which search, watched which video, spent so much -or so little- time on his activities. We believe the ethical consequences are tremendous.
3.5 Situation today
The Moodle plug-in is working and is being tested, both for technical bugs and for UX issues. It is tested on moodle 3.6 and now several evaluations are happening in parallel in other moodle versions (3.7, 3.8, 3.9) with other moodle communities inside the universities and outside. We aim for the different tests to have been completed before the end of 2020. A second version of this tool is under preparation with features allowing more interaction: annotating shared material, using the annotations for better recommendations, proposing real learning trajectories based on the classroom’s experience.
4. Final remarks
There are many challenges for OER. Some of these are technological. Artificial intelligence can -in part- help and provide tools which can not only make it easier to create OER, to find OER and to use OER. It can also provide teachers with new pedagogical tools.
Acknowledgement
This research is conducted as part of the X5GON project (www.x5gon.org) funded from the EU’s Horizon 2020 research and innovation programme grant No 761758 and par-tially funded by the EPSRC Fellowship titled ”Task Based Information Retrieval”, under grant No EP/P024289/1.
Bibliography
[1] Bulathwela, S. and Yilmaz, E. and Shawe-Taylor, J.. Towards Automatic, Scalable Quality Assurance in Open Education}, Workshop on AI and the United Nations SDGs at International Joint Conference on Artificial Intelligence, 2019.
[2] de la Higuera, C., Le Capitaine, H., Romdhane, W. B., Leray, P., & Hernandez, N. X5GON: Vers l’utilisation de l’Intelligence Artificielle pour une meilleure utilisation des Ressources Éducatives Libres. PFIA 2018, Nancy, 2018.
[3] Orlic, D. ,Makarovic, B., Bogataj Jančič, M., Abe Hsuan, A. Ethical Data Management and Data Management Plan Year 2. 2019 https://www.x5gon.org/wp-content/uploads/2019/10/D9.2_Ethical_Data_Management_and_Data_Management_Plan.pdf
[4] Valor Miró, J-D., Baquero-Arnal, P., Civera, J., Turró, C., and Juan, A. Multilingual Videos for MOOCs and OER. Educational Technology & Society, 21(2):1–12, 2018.
[5] Rogers, Y. The New Zeitgeist: Human-AI. Keynote talk at AIE-20. https://aied2020.nees.com.br/