Automatic Classification and Semantic Retrieval for Low-Resource Parliamentary Archives: A Case Study of the DRC Senate

Otshudiakoy Jean; Djungu Ahuka Saint Jean

doi:10.30871/jaic.v10i4.13356

Authors

Otshudiakoy Jean Department of Mathematics, Statistics and Computer Science, Faculty of Science and Technology, University of Kinshasa, Kinshasa, Democratic Republic of the Congo
Djungu Ahuka Saint Jean CRIA-Center for Research in Applied Computing, Kinshasa, Democratic Republic of the Congo; Department of Mathematics, Computer Science and Statistics, University of Kinshasa

DOI:

https://doi.org/10.30871/jaic.v10i4.13356

Keywords:

Automatic Classification, Parliamentary Archives, OCR, TF-IDF, CamemBERT, Probabilistic Fusion, Semantic Search, Low-Resource NLP

Abstract

Parliamentary archives preserve the administrative, legal and political memory of legislative institutions. At the Senate of the Democratic Republic of the Congo, many documents are scanned PDF files, which makes manual indexing, classification and retrieval slow. This study does not propose a new algorithm. Its contribution is the applied integration of automatic classification and hybrid semantic retrieval in a French, OCR-based and low-resource parliamentary archive. The corpus contains 420 documents classified into five categories: orders, decrees, laws, ordinances and minutes. The models include TF-IDF with linear SVM, logistic regression and Naive Bayes, CamemBERT embeddings with logistic regression, and a weighted probabilistic fusion: P_fusion = alpha P_TFIDF + (1-alpha) P_CamemBERT, with alpha = 0.55. On the fixed test set, the hybrid model obtains accuracy = 0.786 and macro F1 = 0.815. Under stratified five-fold cross-validation, TF-IDF + linear SVM is the most stable lexical baseline, with accuracy = 0.821 +/- 0.052 and macro F1 = 0.834 +/- 0.050. These protocols are interpreted separately. For thirteen curated archival queries, hybrid retrieval performs best at k = 10, while TF-IDF obtains the highest MRR. The results show that lexical models are strong for categories with stable legal markers, whereas hybrid retrieval is useful for broader semantic access. The proposed system is a human-validated archival workflow, not a fully autonomous decision tool.

Downloads

Download data is not yet available.

References

[1] C. Sanderson and M. S. Irwin, "Transforming parliamentary libraries: Enhancing processes and services through artificial intelligence," IFLA Journal, 2025.

[2] F. Ariai and G. Demartini, "Natural Language Processing for the Legal Domain: A Survey of Tasks, Datasets, Models, and Challenges," ACM Computing Surveys, 2025.

[3] A. Canning, "Artificial intelligence for reviewing government records and digital records management," Records Management Journal, 2025.

[4] M. A. Hedderich, L. Lange, H. Adel, J. Stroetgen, and D. Klakow, "A Survey on Recent Approaches for Natural Language Processing in Low-Resource Scenarios," NAACL-HLT, 2021.

[5] K. Kowsari, K. Jafari Meimandi, M. Heidarysafa, S. Mendu, L. Barnes, and D. Brown, "Text Classification Algorithms: A Survey," Information, vol. 10, no. 4, 2019.

[6] C. D. Manning, P. Raghavan, and H. Schuetze, Introduction to Information Retrieval. Cambridge University Press, 2008.

[7] N. Reimers and I. Gurevych, "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks," EMNLP-IJCNLP, 2019.

[8] P. Joshi, S. Santy, A. Budhiraja, K. Bali, and M. Choudhury, "The State and Fate of Linguistic Diversity and Inclusion in the NLP World," ACL, 2020.

[9] S. Pakray, A. Gelbukh, and S. Bandyopadhyay, "Natural language processing applications for low-resource languages," Natural Language Processing, 2025.

[10] M. Cueto Aparicio, "Guidelines for AI in Parliaments: best practices and governance principles," IFLAPARL / Westminster Foundation for Democracy, 2025.

[11] O. Q. Osagie, "The usefulness of artificial intelligence to safeguard records in libraries," South African Journal of Libraries and Information Science, 2024.

[12] T. Joachims, "Text Categorization with Support Vector Machines: Learning with Many Relevant Features," European Conference on Machine Learning, 1998.

[13] L. Martin et al., "CamemBERT: A Tasty French Language Model," ACL, 2020.

[14] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding," NAACL-HLT, 2019.

[15] E. Chalkidis, I. Androutsopoulos, and N. Aletras, "Neural Legal Judgment Prediction in English," ACL, 2019.

[16] I. Chalkidis et al., "LexGLUE: A Benchmark Dataset for Legal Language Understanding in English," ACL, 2022.

[17] P. S. Nguyen, V. D. Nguyen, and M. Le Nguyen, "Legal domain knowledge acquisition for low-resource languages," arXiv preprint, 2023.

[18] S. Arimond, M. L. Tran, and M. Grabmair, "Transformer-based legal form classification," arXiv preprint, 2023.

[19] G. Krasadakis, A. Papadakis, and E. Papadakis, "Natural language processing challenges and advances in law consolidation," Electronics, vol. 13, no. 3, 2024.

[20] A. Principe, M. A. Fauzi, and A. D. Mauro, "Long document classification in the transformer era: challenges, models and evaluation," WIREs Data Mining and Knowledge Discovery, 2025.

[21] V. Karpukhin et al., "Dense Passage Retrieval for Open-Domain Question Answering," EMNLP, 2020.

[22] G. Izacard et al., "Unsupervised Dense Information Retrieval with Contrastive Learning," Transactions of Machine Learning Research, 2022.

[23] L. Wang et al., "Text Embeddings by Weakly-Supervised Contrastive Pre-training," arXiv preprint arXiv:2212.03533, 2022.