dzFinNlp at AraFinNLP: Improving Intent Detection in Financial Conversational Agents

TL;DR

The paper tackles multi-dialect intent detection for Arabic financial conversational agents using ArBanking77, a dialect-diverse dataset derived from Banking77 translations. It systematically compares traditional TF-IDF + LinearSVC, LSTM-based models, and transformer-based approaches (XLM-RoBERTa variants and sentence-embedding pipelines) to identify effective strategies for Arabic banking queries. The findings show that a TF-IDF feature union with LinearSVC achieves the strongest development performance ($F1$ up to $93.08\%$), while LSTMs and transformers provide competitive but sometimes lower development scores and notable test-set generalization gaps (reported as $67.21\%$ micro-$F1$ in the abstract). Overall, the work demonstrates the viability of both classic and modern NLP techniques for Arabic financial NLP and suggests avenues for hybrid modeling and domain-specific fine-tuning to improve cross-dialect robustness.

Abstract

In this paper, we present our dzFinNlp team's contribution for intent detection in financial conversational agents, as part of the AraFinNLP shared task. We experimented with various models and feature configurations, including traditional machine learning methods like LinearSVC with TF-IDF, as well as deep learning models like Long Short-Term Memory (LSTM). Additionally, we explored the use of transformer-based models for this task. Our experiments show promising results, with our best model achieving a micro F1-score of 93.02% and 67.21% on the ArBanking77 dataset, in the development and test sets, respectively.