Text Classification for Foreign Trade Documents Diş Ticaret Finansal Metin Siniflandirmasi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636723
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Financial orders, text classification
- Orta Doğu Teknik Üniversitesi Adresli: Evet
Özet
Financial institutions must process a large volume of written customer instructions submitted in various formats. However, these instructions - often provided as free text or semi-structured templates - may contain several sources of noise, such as spelling errors, missing information, and terminological inconsistencies. In this study, we address a multi-class text classification problem using real customer instructions obtained from the foreign trade and foreign exchange transfer operations of a financial institution. The dataset consists of five financial payment categories annotated by domain experts and contains a total of 1,195 documents. Following OCR-based text extraction, we compare Machine Learning, Deep Learning, Large Language Models, and Vision-Language Models. The results show that the Logistic Regression model with a Bag-of-Words (BoW) representation achieves the highest performance with a weighted F1 score of 88.8%, while vision-language models remain around 76.2% weighted F1.