Abstract
The usage of classification systems is a standard method in libraries to organize all kind of materials. The Dewey Decimal Classification System (DDC) is widely used for this task. Even though approaches exist since the 1970s to automate this classification task, it is most often still a time consuming manual process. With the constantly increasing number of publications the need for automation support is growing. Current approaches have certain limitations e.g. only mono- or bi-lingual support, limited accuracy for research domains, limited to higher levels in the DDC hierarchies. The usage of Large Language Models (LLMs) opens new possibilities to support librarians in their work. In this paper we present preliminarily a study to evaluate the usage of BERT to handle a DDC classification task in the linguistic domain. In addition, we analyze the effect of a more condensed representation of full text on the performance of LLMs for this task. Results on multilingual texts are comparable to recent performances on monolingual inputs.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
Other studies report that the accuracy of DDC assignment was the highest for Language and the lowest for Literature [5].
- 2.
An issue reported by several papers lies in the proprietary licenses used by the copyright holders for the classifications, which make them difficult to use freely or access automatically, e.g. by ChatGPT.
- 3.
Differences in experiment design makes it difficult to compare results across projects. However, the accuracy of the classifier assignment largely falls in the bracket 0.4–0.6 [7, 15], with recent models towards the upper limit (e.g. 0.668 in [5]). Results can be improved to 0.7–0.8 by broadening the number of recommended classifiers [2], by using large training set on restricted DDC ranges (0.819 reported in [5], 0.8< in [9]), or human intervention (0.9 in [17]).
- 4.
The complete set of classes found in the data: 400–401, 404, 407, 409–415, 417–418, 420–421, 423, 425, 427, 430–433, 435, 437–440, 447, 450, 460, 465, 469–470, 480, 490–492, 495.
- 5.
The outlined approach is roughly based on concepts presented in the blog post [3].
- 6.
April 2024, also see footnote 7.
- 7.
- 8.
https://www.opus-repository.org/, 2024-04-29.
- 9.
E.g. using a modified version of OPUS or, despite the characterization by the BASE team, a different software altogether.
- 10.
https://api.base-search.net/, 2024-04-29.
- 11.
The datasets and two best models are available for open access at https://dx.doi.org/10.5281/zenodo.12935200 (baseline) and https://dx.doi.org/10.5281/zenodo.12941525 (keyword).
- 12.
- 13.
References
Bodenhamer, J.: The reliability and usability of ChatGPT for library metadata. In: Brick & Click Libraries Conference Proceedings, pp. 1–13. Brick & Click (2023)
Borovič, M., Ojsteršek, M., Strnad, D.: A hybrid approach to recommending universal decimal classification codes for cataloguing in Slovenian digital libraries. IEEE Access 10, 85595–85605 (2022). https://doi.org/10.1109/ACCESS.2022.3198706
Fritze, T.: Selecting subject specific records from the bielefeld academic search engine (part 1). UBLabs Blog (2022). https://labs.ub.uni-frankfurt.de/post/selecting-subject-specific-records-from-the-bielefeld-academic-search-engine-part-1/. Accessed 29 April 2024. blog Post
Golub, K.: Automated subject indexing: an overview. Cataloging Classif. Q. 59(8), 702–719 (2021). https://doi.org/10.1080/01639374.2021.2012311
Golub, K., Suominen, O., Mohammed, A.T., Aagaard, H., Osterman, O.: Automated Dewey decimal classification of Swedish library metadata using Annif software. J. Documentation (2024). https://doi.org/10.1108/JD-01-2022-0026
Grootendorst, M.: KeyBERT: minimal keyword extraction with BERT (2020). https://doi.org/10.5281/zenodo.4461265
Illangarathne, S.: A comparative study of library classification numbers assignment accuracy: openAI language models vs. human classifiers. In: 13th International Conference of the University Librarians Association of Sri Lanka (ICULA - 2023), pp. 83–90. University Librarians Association of Sri Lanka, Colombo (2023)
ISO Central Secretary: Codes arranged alphabetically by alpha-3/iso 639-2 code. Standard, International Organization for Standardization, Geneva, CH (2017). https://www.loc.gov/standards/iso639-2/php/code_list.php
Kragelj, M., Borštnar, M.K.: Automatic classification of older electronic texts into the Universal Decimal Classification-UDC. J. Documentation 77(3), 755–776 (2021). https://doi.org/10.1108/JD-06-2020-0092
Kua, E.: Non-western languages and literatures in the Dewey decimal classification scheme. Libri 54(4), 256–265 (2004). https://doi.org/10.1515/LIBR.2004.256
Lösch, M., Waltinger, U., Horstmann, W., Mehler, A.: Building a DDC-annotated corpus from OAI metadata. J. Digit. Inform. 12(2) (2011)
Mattmann, C.A.: Tika-Python (2023). https://pypi.org/project/tika/
Pedregosa, F., et al.: Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12(Oct), 2825–2830 (2011)
Reimers, N., Gurevych, I.: Sentence-BERT: sentence embeddings using siamese BERT-networks. In: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics (2019). http://arxiv.org/abs/1908.10084
Roy, A., Ghosh, S.: Automated subject identification using the universal decimal classification: the ANN approach. J. Inform. Knowl. 60(2), 69–76 (2023). https://doi.org/10.17821/srels/2023/v60i2/170963
Suominen, O., et al.: Annif (2023). https://doi.org/10.5281/zenodo.8262313, https://annif.org/
Wang, J.: An extensive study on automated dewey decimal classification. J. Am. Soc. Inform. Sci. Technol. 60(11), 2269–2286 (2009). https://doi.org/10.1002/asi.21147
Weber, T.: The curation of language data as a distinct academic activity: a call to action for researchers, educators, funders, and policymakers. J. Open Humanit. Data 7(28) (2021). https://doi.org/10.5334/johd.51
Wolf, T., et al.: Huggingface’s transformers: state-of-the-art natural language processing. CoRR abs/1910.03771 (2019). http://arxiv.org/abs/1910.03771
Acknowledgments
This work was partially funded by the DFG under FID Linguistik (326024153).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2024 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Ho, C.W.C., Weber, T., Fritze, T., Risse, T. (2024). Towards Multilingual LLM-Based Approaches for Automatic Dewey Decimal Classification. In: Antonacopoulos, A., et al. Linking Theory and Practice of Digital Libraries. TPDL 2024. Lecture Notes in Computer Science, vol 15178. Springer, Cham. https://doi.org/10.1007/978-3-031-72440-4_3
Download citation
DOI: https://doi.org/10.1007/978-3-031-72440-4_3
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-72439-8
Online ISBN: 978-3-031-72440-4
eBook Packages: Computer ScienceComputer Science (R0)