GitHub R-CMD-check CRAN CRAN status Downloads Project Status: Active - The project has reached a stable, usable state and is being actively developed.

The R package Artificial Intelligence for Education (aifeducation) is designed for the special requirements of educators, educational researchers, and social researchers. The target audience of this package are educators and researchers with no coding skills who would like to develop their own models, as well as people who would like to use those models created by other researchers/educators. The package supports the application of Artificial Intelligence (AI) for Natural Language Processing tasks such as text embedding and classification under the special conditions of the educational and social sciences.

Features Overview

Currently, the package focuses on classification tasks which can either be used to diagnose characteristics of learners from written material or to estimate the properties of learning and teaching material. In the future, more tasks will be implemented.


You can install the latest stable version of the package from CRAN with:

#Minimal version

#Full version

You can install the development version of aifeducation from GitHub with:

#Minimal version
                         dependencies = "Imports")
#Maximal version
                         dependencies = TRUE)

The minimal version includes all functions but is limited to the use of transformers. The full version additionally includes Aifeducation Studio (graphical user interface) and older approaches (GlobalVectors, Topic Modeling).

Further instructions for installation can be found in vignette 01 Get Started.

Please note that an update of your version of aifeducation may require an update of your python libraries. Refer to 01 Get Started for more details.

Graphical User Interface Aifeducation Studio

The package ships with a shiny app that serves as a graphical user interface.

Figure 1: Aifeducation Studio

Aifeducation Studio allows users to easily develop, train, apply, document, and analyse AI models without any coding skills. See the corresponding vignette for more details: 02a Using Aifeducation Studio.


Training AI models consumes time and energy. To help researchers estimate the ecological impact of their work, a sustainability tracker is implemented. It is based on the python library ‘codecarbon’ by Courty et al. (2023). This tracker allows to estimate the energy consumption for CPUs, GPUs and RAM during training and derives a value for CO2 emission. This value is based on the energy mix in the country where the computer is located.

PyTorch and Tensorflow Compatibility

This package allows all supported models either based on ‘PyTorch’ or ‘tensorflow’, thus providing a high level of flexibility. Even pre-trained models can be used with both frameworks in some cases. The following table provides more details:

Table: Framework compatibility

Model PyTorch tensorflow Weight Sharing
BERT Yes Yes Yes
RoBERTa Yes Yes Yes
DeBERTa Yes Yes Yes
Funnel Transformer Yes Yes Yes
Longformer Yes Yes Yes
Text Embedding Classifier Yes Yes No

Please not that tensorflow is currently supported for the following versions: 2.13-2.15.

Classification Tasks

Transforming Texts into Numbers

Classification tasks require the transformation of raw texts into a representation with numbers. For this step, aifeducation supports both newer approaches such as BERT (Devlin et al. 2019), RoBERTa (Liu et al. 2019), DeBERTa version 2 (He et al. 2020), Funnel-Transformer (Dai et al. 2020), and Longformer (Beltagy, Peters & Cohan 2020) and older approaches such as GlobalVectors (Pennington, Socher & Manning 2014) or Latent Dirichlet Allocation/Topic Modeling in classification tasks.

aifeducation supports the use of pre-trained transformer models provided by Hugging Face and the creation of new transformers, allowing educators and researchers to develop specialized and domain-specific models.

The package supports the analysis of long texts. Depending on the method, long texts are transformed into vectors at once or, if too long, are split into several chunks which results in a sequence of vectors.

Training AI under Challenging Conditions

For the second step within a classification task, aifeducation integrates some important statistical and mathematical methods for dealing with the main challenges in educational and social sciences for applying AI. These are:

In order to deal with the problem of imbalanced data sets, the package integrates the Synthetic Minority Oversampling Technique into the learning process. Currently, the Basic Synthetic Minority Oversampling Technique (Chawla et al. 2002), Density-Based Synthetic Minority Oversampling Technique (Bunkhumpornpat, Sinapiromsaran & Lursinsap 2012), and Adaptive Synthetic Sampling Approach for Imbalanced Learning (Hem Garcia & Li 2008) are implemented via the R package smotefamiliy.

In order to address the problem of small data sets, training loops of AI integrate pseudo-labeling (e.g., Lee 2013). Pseudo-labeling is a technique which can be used for supervised learning. More specifically, educators and researchers rate a part of a data set and train AI with this very part. The remainder of the data is not processed by humans. Instead, AI uses this part of data to learn on its own. Thus, educators and researchers only have to provide additional data for the AI’s learning process without coding it themselves. This offers the possibility to add more data to the training process and to reduce labor costs.

Evaluating Performance

Classification tasks in machine learning are comparable to the empirical method of content analysis from the social sciences. This method looks back on a long research tradition and an ongoing discussion on how to evaluate the reliability and validity of generated data. In order to provide a link to this research tradition and to provide educators as well as educational and social researchers with performance measures they are more familiar with, every AI trained with this package is evaluated with the following measures and concepts:

In addition the some traditional measures from the machine learning literature are also available:

Sharing Trained AI

Since the package is based on torch, tensorflow, and the transformer libraries, every trained AI can be shared with other educators and researchers. The package supports an easy use of pre-trained AI within R, but also provides the possibility to export trained AI to other environments.

Using a pre-trained AI for classification only requires the classifier and the corresponding text embedding model. Use Aifeducation Studio or just load both to R and start predictions. Vignette 02a Using Aifeducation Studio describes how to use the user interface. Vignette 02b Classification Tasks describes how to save and load the objects with R syntax. In vignette 03 Sharing and Using Trained AI/Models you can find a detailed guide on how to document and share your models.

Tutorial and Guides


Beltagy, I., Peters, M. E., & Cohan, A. (2020). Longformer: The Long-Document Transformer.

Berding, F., & Pargmann, J. (2022). Iota Reliability Concept of the Second Generation. Berlin: Logos.

Berding, F., Riebenbauer, E., Stütz, S., Jahncke, H., Slopinski, A., & Rebmann, K. (2022). Performance and Configuration of Artificial Intelligence in Educational Settings.: Introducing a New Reliability Concept Based on Content Analysis. Frontiers in Education, 1-21.

Bloemen, A. (2011). Lernaufgaben in Schulbüchern der Wirtschaftslehre: Analyse, Konstruktion und Evaluation von Lernaufgaben für die Lernfelder industrieller Geschäftsprozesse. Hampp.

Bunkhumpornpat, C., Sinapiromsaran, K., & Lursinsap, C. (2012). DBSMOTE: Density-Based Synthetic Minority Over-sampling Technique. Applied Intelligence, 36(3), 664–684.

Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: Synthetic Minority Over-sampling Technique. Journal of Artificial Intelligence Research, 16, 321–357.

Courty, B., Schmidt, V., Goyal-Kamal, Coutarel, M., Feld, B., Lecourt, J., & … (2023). mlco2/codecarbon: v2.2.7.

Dai, Z., Lai, G., Yang, Y. & Le, Q. V. (2020). Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing.

Devlin, J., Chang, M.‑W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In J. Burstein, C. Doran, & T. Solorio (Eds.), Proceedings of the 2019 Conference of the North (pp. 4171–4186). Association for Computational Linguistics.

Gruber, N., & Kreuzpointner, L. (2013). Measuring the reliability of picture story exercises like the TAT. PloS One, 8(11), e79450.

Gwet, K. L. (2014). Handbook of inter-rater reliability: The definitive guide to measuring the extent of agreement among raters (Fourth edition). STATAXIS.

Haixiang, G., Yijing, L., Shang, J., Mingyun, G., Yuanyue, H., & Bing, G. (2017). Learning from class-imbalanced data: Review of methods and applications. Expert Systems with Applications, 73, 220–239.

He, H., Bai, Y., Garcia, E. A., & Li, S. (2008). ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence) (pp. 1322–1328). IEEE.

He, P., Liu, X., Gao, J. & Chen, W. (2020). DeBERTa: Decoding-enhanced BERT with Disentangled Attention.

Krippendorff, K. (2019). Content Analysis: An Introduction to Its Methodology (4th Ed.). SAGE.

Lee, D.‑H. (2013). Pseudo-Label: The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Networks. CML 2013 Workshop: Challenges in Representation Learning.

Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019). RoBERTa: A Robustly Optimized BERT Pretraining Approach.

Luan, H., Geczy, P., Lai, H., Gobert, J., Yang, S. J. H., Ogata, H., Baltes, J., Guerra, R., Li, P., & Tsai, C.‑C. (2020). Challenges and Future Directions of Big Data and Artificial Intelligence in Education. Frontiers in Psychology, 11, 1–11.

Pennington, J., Socher, R., & Manning, C. D. (2014). GloVe: Global Vectors for Word Representation. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing.

Stütz, S., Berding, F., Reincke, S., & Scheper, L. (2022). Characteristics of learning tasks in accounting textbooks: an AI assisted analysis. Empirical Research in Vocational Education and Training, 14(1).

Wong, J., Baars, M., Koning, B. B. de, van der Zee, T., Davis, D., Khalil, M., Houben, G.‑J., & Paas, F. (2019). Educational Theories and Learning Analytics: From Data to Knowledge. In D. Ifenthaler, D.-K. Mah, & J. Y.-K. Yau (Eds.), Utilizing Learning Analytics to Support Study Success (pp. 3–25). Springer.