Abstract:

At the University of Applied Science Northwestern Switzerland (FHNW) it can be observed that the usage of preferred learning material shifted in the past years from books and articles to videos, summaries and flashcards. The reasons for that are mainly the learning habits which changed during the Covid-19 pandemic, a gradual substitution of books and articles by slides and master solutions for exams in e-learning environments and the increase of part-time students which need to optimally organize themselves in order to manage work and study. Overall it can be observed that there is an increasing demand for on-hands learning material which allows to "consume" the needed learning material in short time. In order to cope with the changing environment students need more and more tools which support the creations of "summaries" for arbitrary domains, i.e. module content. This paper describes an AI-based approach to generate flashcards for arbitrary domains. Flashcards are a very popular aid for supporting systematic learning which support to memorize mainly-fact-based content with pairs of questions and answers.

In a specialization module of the programm BSc Business Information Technology (BIT) at the FHNW a Python-based prototype was created for generating flashcards. A PDF file containing essential learning material is turned into an image and serialized into text using the OCR software Tesseract. The extracted information is sent to OpenAI as prompts via the OpenAI Assistant. OpenAI returns flashcard questions and answers which are combined into JSON objects. Based on the chosen output format, flashcards are saved as JSON/CSV or automatically converted into an ANKI (free flashcard software) deck.

First results are very promising regarding the quality of generated flashcards. Currently in a second round this prototype will be be improved. Main areas of investigation are mainly the substitution of OpenAI with an on-premise LLM solution, the further training of said LLM and the evaluation of enriching OCR with postprocessing facility for Text Mining. An on-premise alternative to OpenAI is necessary to speed up training and to be prepared for future scenarios with critical or confidential data (i.e. in-house training, critical domains like medicine, IT-security, military projects, etc.). The investigation of Text Mining is necessary to find out if filtering and structuring the data (i.e. eliminating stopwords, etc.) before sending it to the LLM, leads to an improved quality of flashcards.

The proposed approach can be used for arbitrary domains and will be further tested and enhanced in the upcoming months. In a first step a typical module of the BIT programme is investigated with all relevant material as input (i.e. books, slides, mock-up and real exams, cheatsheets, etc.) which are available on the e-learning platform Moodle. In a second step the goal is to offer the flashcard software for all programms of the School of Business at the FHNW (all Bachelor and Master programs, courses in further education).

Keywords:

Flashcard, Artificial Intelligence, Python, self-learning.