Leveraging Social Media and Multitask Natural Language Processing to Detect Financial Consumer Complaints

The widespread reach and one-to-one interactions with businesses through social media now allow people to speak their minds easily.It enhances customer services by making it easier for users to raise complaints and hold the businesses accountable, while also allowing companies to respond promptly in real time.Public visibility can also support others with similar issues and influence other potential customers.This benefits both customers and businesses, creating a feedback loop that can lead to improvements in products or services.In natural language processing, classifying complaints based on severity level is a key challenge.Also, in human conversations, a single sentence can carry multiple meanings, making it a complex task to classify these meanings effectively using a multitasking model.In this work, we proposed a multitasking model applied to a financial dataset containing tweets from bank customers.The proposed multitasking framework classifies four tasks: complaint label, severity label, emotion, and sentiment for each tweet.Our model outperformed existing models, achieving an accuracy of 97.93% and an F1 score of 97.95% for the complaint identification task, and an accuracy of 84.49% with an F1 score of 84.02% for the complaint severity task.