Evonik Digital Research

A joint research project of EVONIK and the University of Duisburg-Essen

Intelligent contextualized information access for learning

Access to information and knowledge bases is one of the key benefits from a successful digital transformation. However, learners are often lacking the abilities to access and to filter the right information. Methods of Artificial Intelligence (AI) have the potential to support learners by providing intelligent information access based on the learners' context.

In the envisioned scenario, learning resource recommendation is the most prominent type of intelligent information access. Such resources are either Open Educational Resources (OER), course materials inside the learning environment, or external materials from a private repository of EVONIK ("Mediathek"). A prototype for the provision of information access through AI has been provided in several modalities. For example, it has been embedded into the learning environment for video-based learning, where the information access is provided based on automatically extracted keywords that identify the context of the learner inside this course. Therefore, resources can be linked and discovered automatically using techniques of AI.

The prototype of this mechanism is an app that handles three different media types as an input: text, audio, and video. For the input the app automatically extracts keywords that represent it's content. In the case of video, audio is extracted, segmented and automatically transcribed using speech-to-text service APIs. Afterwards, keywords are extracted automatically from the medium. The learning objects of the learning scenario can be used as an input in order to span the context for the information access. This counts as well for active learning scenarios, where learners produce content themselves. The extraction itself is based on a tf-idf model and DBPedia Spotlight, which renders the results as terms that exist in Wikipedia. Furthermore, the extracted concepts are passed to search engines (i.e., Chemgapedia, EVONIK Mediathek, Google) in order to propose suitable learning resources. Due to privacy restrictions, this does not include a personal trajectory or history of learners.