This model is a fine-tuned version of RoBERTa-large [1]. It was trained on 1,400 employee reviews to measure corporate culture. More specifically, it predicts which of the four culture dimensions of the Competing Values Framework [2,3] best fits the text at hands, i.e., what is the dominant culture.
The model assigns one of four possible labels:
0 (clan): Text is best described by a clan (collaborating) culture.
1 (adhocracy): Text is best described by an adhocracy (creating) culture.
2 (market): Text is best described by a market (competing) culture.
3 (hierarchy): Text is best described by a hierarchy (controlling) culture.
For details on the model and its performance, see Koch and Pasch (2023). Please cite this article when using the model:
S. Koch and S. Pasch, "CultureBERT: Measuring Corporate Culture With Transformer-Based Language
Models," in 2023 IEEE International Conference on Big Data (BigData), pp. 3176-3184. doi: 10.1109/BigData59044.2023.10386765
Please see the following
tutorial on how to apply CultureBERT to measure corporate culture in your own text documents:
https://github.com/Stefan-Pasch/CultureBERTOther References:
[1] Liu, Y.; Ott, M.; Goyal, N.; Du, J.; Joshi, M.; Chen, D.; ... & Stoyanov, V. (2019). Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
[2] Cameron, Kim S.; Quinn, Robert E. (2011): Diagnosing and Changing Organizational Culture. Based on the Competing Values Framework. 3rd ed. San Francisco (CA): Jossey-Bass.
[3] Quinn, Robert E.; Rohrbaugh, John (1983): A Spatial Model of Effectiveness Criteria: Towards a Competing Values Approach to Organizational Analysis. In Management Science 29 (3), pp. 363–377. DOI: 10.1287/mnsc.29.3.363.
