Let the Model Make Financial Senses: A Text2Text Generative Approach for Financial Complaint Identification.

A complaint frequently expresses the complainer’s dissatisfaction or objectionable notion to support a belief or claim against a party or parties. Financial loss, material inconvenience, and distress are sufficient examples to intensify the need for an automated complaint analysis tool in the financial domain, particularly on social media with diverse information-related affairs. Recently, advanced approaches like complaint detection with machine learning have escalated the research interest in the area of natural language processing. Earlier, the only research focus was on how complaints are identified linguistically. Substantial modern complaint analytical models attempt to bridge the gap between the interpretability and explainability of financial complaint detection tasks. To address this, we extend an existing complaint dataset X-FINCORP, with the rationale or cause annotations for the complaint/non-compliant labels. Each instance in the dataset is now associated with five labels: complaint, emotion, polarity, severity, and rationales. Our proposed model addresses the multi-task problem as a text-to-text generation task by utilizing a generative framework. Additionally, we introduce commonsense as external information to draw more informative intuitions and enhance the overall performance of the proposed generative model. The empirical results validate the generality of our proposed model over several evaluation metrics compared to state-of-the-art models and other baselines (Resources available at https://github.com/appy1608/Financial-Complaint-Identification .).