Hilbot-FI Dataset: A Low-Resource Financial Intent Classification Dataset

The Hilbot-FI Dataset is a low-resource financial intent classification dataset designed for evaluating NLP models under realistic short-text and class-imbalanced conditions.

The dataset contains 1,525 total samples across 33 intent labels, with predefined train/test splits of 1,220 training samples and 305 test samples. The processed vocabulary contains 542 unique tokens.

The dataset combines structured financial-record-derived queries with conversational financial intent patterns, reflecting the mixed nature of real-world financial assistant inputs. It is intended for research on financial conversational systems, decision-support chatbots, class imbalance, short-text classification, and data-centric evaluation of classical, neural, hybrid, and transformer-based NLP models.

The release includes the complete dataset, predefined train/test splits, intent patterns, a data dictionary, label definitions, citation metadata, and license information.