#	Pagina
attuale pagina	/open-h2020/projects/211074/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - X5gon (X5gon: Cross Modal, Cross Cultural, Cross Lingual, Cross Domain, and Cross Site Global OER Network)

Teaser

Summary

X5GON is an EU project (2017â€“2020) with eight partners from the UK, France, Slovenia, Germany and Spain. It is intended to help students and the general public learn effectively and enjoyably by providing a personalized route through appropriately prioritized Open Educational Resources (OER).

Our plan is to develop an extensive architecture, where state-of-the-art machine learning and recommender algorithms are deployed to crawl, classify and understand these resources so that we can then determine how best to help people learn in a way most suited to them. The X5GON project therefore aims at harvesting OER data and creating the first AI powered platform for OERs that will allow teachers and students, businesses and educational institutions to access OER from everywhere at any time in various formats such as video, text or pictures, different topics and languages.

We aim to bring the power of AI via machine learning models to automatically determine the quality of OER, automatically translate OER and automatically profile, understand and reference OERS to learners from different countries and cultures, and thus allowing learners to easily bridge educational barriers on many levels. We will do all of the above in compliance with ethics and privacy issues in relation to all aspects of the project.

The goal of the project is to create a platform for conversion of scattered OERs available in various modalities across Europe and the globe. The vision is therefore to overcome the fragmentation of all OER sites and break accessibility barriers and provide end users with a common learning experience.

To achieve its goal, the project pursues the following challenging scientific and technological objectives:

1. Cross-modal: develop technologies for multimodal content understanding
2. Cross-site: develop technologies to transparently accompany and analyse across sites
3. Cross-domain: develop technologies for cross domain content analytics
4. Cross-language: develop technologies for cross lingual content recommendation
5. Cross-cultural: develop technologies for cross cultural learning personalisation.

The project will create three services X5oerfeed, X5analytics and X5recommend and run a series of pilot case studies that enable the measurement of the broader goals of delivering a useful and enjoyable educational experience to learners in different domains, at different levels and from different cultures. Two exploitation scenarios are planned: (i) free use of services for OER, (ii) commercial exploitation of the multimodal, big data, real-time analytics pipeline.

Work performed

1. Learning rich content representations is part of the Cross-modal technologies for multimodal content understanding.

A major challenge in the project and in general in the field of OER, is how to automatically learn representations of OERs, and hence address the quality and authority of each OER, as well as the topics covered by the OER. In solving this we have carried out four major tasks, namely (i) we created a stable dataset of OERs, (ii) developed accurate quality assurance models, (iii) evaluated and interpreted these quality models, and (iv) started developing initial content representations. Amongst numerous machine learning unsupervised and semi-supervised methods, we found Wikification to be a promising technique to identify the topics in lectures. The result is a built model that can currently assess quality with 71% accuracy.

2. Analytics Infrastructure, Services and API is part of the Cross-modal and Cross-site technologies to transparently accompany and analyse users across sites and Cross-domain technologies for cross domain content analytics.
One of the core objectives of X5GON was to create a sustainable and scalable software platform able to sustain a large input of multimodal content, comprised of text, video and audio OER. The system now connects several technical components such as the OER resources crawler, pre-processing pipeline, recommendation engine, and monitoring system. The current version of the platform has ingested 80.000 OER and can link the different components of the system and perform analytical functions.

3. Learning Analytics Engine and Recommendation Engine are part of the Cross-language and Cross-cultural aspects for cross lingual content recommendation, and cross-cultural learning personalisation.

The initial Learning Analytics engine and its API are developed with data from the OERs at the X5GON pilot sites, namely from (JSI) a collection of 17972 videos over 8 languages, all transcribed and, where necessary, translated, (UOS) 276 videos in German (DE), transcribed and translated to English (EN), and (UPV) with 4166 videos in Spanish (ES) and Catalan (CA), translated to EN. The main features are an extraction of metadata, a graphical presentation of a resource and related resources which takes into account the difficulty of a resource and its length. These rely on improved language models and translation models and include features such as the computation of the difficulty of a text, the capacity of identifying â€œmissingâ€ lectures, etc.

4. Piloting and Studies in the wild are part of the Cross-language and Cross-cultural aspects for cross-lingual content recommendation, and cross-cultural learning personalisation.

An essential challenge in X5GON is to understand the experiences of our target groups, revealing the factors that hold user engagement and what makes learning enjoyable and rewarding. The non-technical piloting, in-the-wild research and design, resulted in findings from in-situ OER user observations, a series of learner-centric interface designs and evaluations, in-the-wild testing of a novel mechanism for peer support in OER, and a conceptual framework to guide the exploration and evaluation of novel and existing OER use cases.

5. Sustainability and Scalability

The project aims at ensuring a long-lasting, sustainable impact in Europe and beyond, by building on the platform and services developed during the X5GON project. Together, these form the framework of technological, societal and research excellence for our sustained offer and impact beyond the project end by (i) creating a value proposition for the introduction of Artificial Intelligence methods in the OER space, and (ii) translating these into a value proposition for policy makers, industry, researchers, academia and civil society. We specifically work on 2 exploitation scenarios with initial business streams in the open and commercial, with 7 defined business cases and 17 initial business models.

Final results

We have identified 4 categories of expected impact to which X5GON will make a potential contribution. The current progress of the project and the business analysis performed have confirmed our expectations of the impact the project will have in all 4 categories: Learning rich content representations: automatic understanding of OER quality; analytics Infrastructure, Services and API: creating an AI powered OER platform; Learning Analytics Engine and Recommendation Engine: breaking the language and accessibility barrier to OER; and Data-driven policy: Governmental validation of X5GON platform.