👨‍🎓Learning analytics to personalize education

Personalized learning is considered to be the most efficient of all training and education approaches because it is tailored to the individual needs, interests and learning style of each student or learner.

Personalized learning allows for the development of customized learning plans and activities that are based on the student's abilities, interests, and goals. This approach allows for a more efficient use of time and resources, as students are able to progress at their own pace, focusing on the areas where they need the most help and skipping over material they already know.

However personalized education in current education and training organizations requires a high ratio of teacher/trainer per learner which makes it expensive and limits its democratization to a wider population. Additionally, the direct observation of a pedagogical activity in real time by a teacher alters the way a learner actually works and usually biases the way the activity is performed in terms of exercising different approaches and learning from errors that limits the impact of the activity.

Recent advances in AI and automatic data collection techniques are paving the way for personalized learning at lower cost and thus large scale. Real time data collection is furthermore allowing a very flexible way to propose either “in activity” guidelines to the learner or provide delayed recommendations proposing a more global message from the whole educational activity. For this to happen, it is critical to have a trusted dataspace of services to collect, store and process the learning records of learners. These learning records are the elementary data that will fuel recommendation engines, dashboard visualizations, and other related services to provide more personalized and individualized learning experience to all learners within Europe, whether students in school or workers in SMEs needing reskilling.

Many actors, both public and private (edtech tools, universities, schools, high-schools, employers, training organizations, edtech tools) are working to improve the collection, storing and processing of learning records and their uses in learning analytics. This must be done in a way that allows either to store their data in their own infrastructures or in a dedicated secured and sovereign cloud solution within a framework that guarantees their interoperability with other information systems. Indeed, it is very interesting to provide information beyond the pedagogical activity data to better guide learners: past curriculum is a tremendous source of information to understand the possible reasons of the deficiencies of a student and propose corrective activities, current curriculum and its description in terms of competences can be important to better associate a given activity output, possibly transversal to different educational programs such as a quiz application for instance, with the objectives of the current training of that learner, etc.

All these stakeholders function in silo are currently with no efficient way of interconnecting their services and access each other’s data:

• To this day, a EU citizen does not have an easy way to access all their own learning traces from school, middle school, high school, university, higher education or professional training. There could be no Lifelong Learning for EU citizens if there is no Lifelong availability and portability of learning data.

• By lack of good practice or knowledge of regulations (GDPR), learning records may often contain personal data, their use and sharing among organizations has been limited. Still there are technical solutions for protecting the personal data in the learning records that could enable data exchange and full GDPR compliance.

• While many training organizations are aware of the potential of personalized learning, most adaptive learning systems available today have been trained on limited data sets. Because these data sets are too small, they may contain bias and have limited range of application. By combining data sets of learning records of a few organizations, we could highly improve the efficiency and inclusivity of personalized learning systems.

• Most training organizations are better at storing physical - paper and pen- learning records than digital learning records. This is quite paradoxical as collecting digital learning records is less expensive and easier to automatize than physical learning records.

• While standards have been slowly emerging (i.e. xAPI), the data model of Learning records may widely differ in most organizations. This technical barrier comes in addition to the legal barrier.

• The learning activities are evolving fast and have many new forms. Yesterday, learning was mainly happening in the physical classroom or with books. Now, learning activities encompass a wide array of physical activities outside the classroom (in museum, directly in the workplace, project-based learning, peer-to-peer learning) and also an even wider array of digital activities (in LMS/eLearning, videos, virtual classrooms, web browsing, apps, podcasts, chats, etc). Tomorrow, new learning activities will emerge such as virtual reality (VR), augmented reality (AR) or other immersive environments (metaverse).

• Current Learning analytics approaches should evolve to be able to collect a wider array of learning records and to access and process them from a large number of organizations. A European dataspace as it will be established by this project will help to evolve from siloted learning analytics with limited impact to Distributed Learning Analytics with a much wider impact.

This use case will provide people and organizations with a set of tools to easily implement Distributed Learning Analytics:

• People will have Lifelong availability of their Learning Records, even if their data is held by many different training organizations or employers using many different data models.

• People will be able to easily import, export, store, share, consent and control the access to their Learning Records.

• AI providers will be able to train their models on larger amounts of Learning records, without contractual, legal or privacy issues and with a fair distribution of value.

• Data providers of Learning records (e.g. training organizations or edtechs) will be able to easily share their data and improve their own services, without requiring expensive data preparation (data model alignment/mapping) or equally expensive ad-hoc contractualization.

• Training organizations will be able to offer personalized learning at lower cost and thus better tailor their learning offer to the needs of their learners.

• Training organizations will improve the inclusivity of their training offers.

• Learning activities happening outside the classrooms will be better integrated and enable a 360° view of learning activities of each learner.

Stakeholders

In more detail, here is the value proposition of these tools for the different stakeholder

People:

• get personalized learning experiences

• get lifelong access and portability of their learning records

• get 360° view of all their learning activities (both physical and digital)

• can share their full learning records with relevant stakeholders

Universities / training organizations:

• can contribute their learning records

• can individualize their training offer for each learner

• can personalize the learning experience of each learner

• get value from their learning records instead of being a cost (hosting)

Employers:

• Get better return on investment (ROI) from their training spending

• Get worker with the right skills required (instead of general training)

• provide employees with innovative upskilling and career mobility services

• get precise education and training background of employee

High schools

• Better detect students who need specific help (e.g. dyslexia)

• get 360° view of each student learning activities (both physical and digital)

Edtechs / AI Providers:

• provide more easily their services and their data to the ecosystem

• get more users and clients

• provider better personalized services thanks to better data access

Infrastructure providers:

• provide services and building blocks to enable data sharing (consent, contract, interoperability, data visualization, decentralized processing, etc)

• get organizations to use their services

Orchestrator:

• provide the ecosystem portal

• coordinate governance, use cases and business model discussions in the ecosystem

• get part of the value generated by the ecosystem through commissions, fees, etc

Prometheus-X gathers numerous data and AI providers that can make these use cases happen:

Users	AI Providers	Data providers
Académie de Nancy: public body managing 126 high school	Inokufu: edtech company providing API access to learning object recommendation engine and competency matching.	ANT: More than 27M xAPI learning traces generated from learner activity
Académie de Rennes: public body managing 112 high schools	Prof en poche: speech Recognition of children in classroom, Handwritten digit recognition, object detection and segmentation
Maskott:Edtech company providing learning content for K12 schools with 20 000 schools, 106 000 teachers with learning traces of est. 1 million students	EvidenceB: provide an API to evaluate the level of student over a graph of activities, estimate attentional disorder or knowledge gap	WEANLY (Oktonine): -10 000 learner profiles years of study, domain, specialization, activities, list of to-be-acquired competencies) list of performed activities, list of validated competencies
EvidenceB (EB): edtech company providing learning apps to K12 students, skills matching learning analytics. 500k users in 10 countries	Cabrilog: adaptive learning solution for K-12 mathematics training courses	EDUNAO: Edunao deliver managed Moodle environments to hundreds of organisations
Le Cnam: high. Ed. organization dedicated to lifelong professional training, with more than 200 training centers in France and 50k+users	Maskott: providing an AI-based content recommendation system for teachers, encouraging students to form their communities with students that share similar interests	Inokufu: dataset of learning traces (xAPI) for 2k users in transition to high demand job
Antares (ANT): company, providing 75% of high schools in Germany with LMS		UoK: student data from 10 000 students

Building blocks mobilized:

In order to make such a use case a reality, different building blocks are needed. Prometheus-X is developing them:

• Consent: to enable people to control their data between all parties in a human-centric way

  • consent agent: to suggest to people the apps / organisations that best fit their needs

• Contract: to ensure trust and compliance between organisations sharing data

• Identity: to ensure authentification of orgs and people sharing data

• Learning traces interop: to allow the translation of learning profiles into xAPI, to ensure interoperability between the different repositories and data models concerning learning traces

  • AI metadata enrichment: algorithms capable of extracting from raw data the activity that allowed the acquisition of competencies on the basis of common reference systems (ROME/ESCO/RECTEC...);

  • Learning Object Metadata crowd tagging: method of tagging and describing digital learning resources such as videos, presentations, and documents using a crowd of individuals

  • Web Analytics Learning Records Universal Connector: allows for the integration of web analytics data with a Learning Record Store (LRS) using the xAPI (Experience API) standard. It enables the conversion of analytics data, such as the data collected by Matomo, into an xAPI format that can be stored and tracked in an LRS

• distributed data visualisation: to allow results of matchings (courses, exercises, ressources) to be shown wherever the user is

• Personal learning record store: type of cloud-based service that allows individuals to store and manage their own learning records in a central location