@article{WulffBuschhueterWestphaletal.2022,
  author    = {Wulff, Peter and Buschh{\"u}ter, David and Westphal, Andrea and Mientus, Lukas and Nowak, Anna and Borowski, Andreas},
  title     = {Bridging the gap between qualitative and quantitative assessment in science education research with machine learning},
  series = {Journal of science education and technology},
  volume    = {31},
  journal   = {Journal of science education and technology},
  number    = {4},
  publisher = {Springer},
  address   = {Dordrecht},
  issn      = {1059-0145},
  doi       = {10.1007/s10956-022-09969-w},
  pages     = {490 -- 513},
  year      = {2022},
  abstract  = {Science education researchers typically face a trade-off between more quantitatively oriented confirmatory testing of hypotheses, or more qualitatively oriented exploration of novel hypotheses. More recently, open-ended, constructed response items were used to combine both approaches and advance assessment of complex science-related skills and competencies. For example, research in assessing science teachers' noticing and attention to classroom events benefitted from more open-ended response formats because teachers can present their own accounts. Then, open-ended responses are typically analyzed with some form of content analysis. However, language is noisy, ambiguous, and unsegmented and thus open-ended, constructed responses are complex to analyze. Uncovering patterns in these responses would benefit from more principled and systematic analysis tools. Consequently, computer-based methods with the help of machine learning and natural language processing were argued to be promising means to enhance assessment of noticing skills with constructed response formats. In particular, pretrained language models recently advanced the study of linguistic phenomena and thus could well advance assessment of complex constructs through constructed response items. This study examines potentials and challenges of a pretrained language model-based clustering approach to assess preservice physics teachers' attention to classroom events as elicited through open-ended written descriptions. It was examined to what extent the clustering approach could identify meaningful patterns in the constructed responses, and in what ways textual organization of the responses could be analyzed with the clusters. Preservice physics teachers (N = 75) were instructed to describe a standardized, video-recorded teaching situation in physics. The clustering approach was used to group related sentences. Results indicate that the pretrained language model-based clustering approach yields well-interpretable, specific, and robust clusters, which could be mapped to physics-specific and more general contents. Furthermore, the clusters facilitate advanced analysis of the textual organization of the constructed responses. Hence, we argue that machine learning and natural language processing provide science education researchers means to combine exploratory capabilities of qualitative research methods with the systematicity of quantitative methods.},
  language  = {en}
}
@article{WulffMientusNowaketal.2022,
  author    = {Wulff, Peter and Mientus, Lukas and Nowak, Anna and Borowski, Andreas},
  title     = {Utilizing a pretrained language model (BERT) to classify preservice physics teachers' written reflections},
  series = {International journal of artificial intelligence in education},
  journal   = {International journal of artificial intelligence in education},
  number    = {33},
  publisher = {Springer},
  address   = {New York},
  issn      = {1560-4292},
  doi       = {10.1007/s40593-022-00290-6},
  pages     = {439 -- 466},
  year      = {2022},
  abstract  = {Computer-based analysis of preservice teachers' written reflections could enable educational scholars to design personalized and scalable intervention measures to support reflective writing. Algorithms and technologies in the domain of research related to artificial intelligence have been found to be useful in many tasks related to reflective writing analytics such as classification of text segments. However, mostly shallow learning algorithms have been employed so far. This study explores to what extent deep learning approaches can improve classification performance for segments of written reflections. To do so, a pretrained language model (BERT) was utilized to classify segments of preservice physics teachers' written reflections according to elements in a reflection-supporting model. Since BERT has been found to advance performance in many tasks, it was hypothesized to enhance classification performance for written reflections as well. We also compared the performance of BERT with other deep learning architectures and examined conditions for best performance. We found that BERT outperformed the other deep learning architectures and previously reported performances with shallow learning algorithms for classification of segments of reflective writing. BERT starts to outperform the other models when trained on about 20 to 30\% of the training data. Furthermore, attribution analyses for inputs yielded insights into important features for BERT's classification decisions. Our study indicates that pretrained language models such as BERT can boost performance for language-related tasks in educational contexts such as classification.},
  language  = {en}
}