Bank Customers Compliment Analysis by Deep Learning Based Sentiment Analysis and Topic Modelling

Majid Iranpour Mobarakeh; Ali Sebti; Moslem Mohammadi-jenghera; and Ali Ghanbari Sorkhi

Today, the rapid increase in the volume of data, especially in cyberspace and social media, has made organizations seek to improve organizational processes using data analysis. Considering the intense competitive environment, customer satisfaction is considered the main capital of businesses. Therefore, the collection and analysis of customers' opinions has received much attention from researchers. Due to the fact that a large part of customer opinions is available in text form in social networks and web systems, sentiment analysis or opinion analysis is presented as a solution to evaluate customer opinions, which with text analysis automatically discovers opinions and opinions about an Entity pays. As one of the banking service providers, Mehr-e-Iran Bank receives customer complaints in text form through the online complaint registration system. In this work, the goal is to analyze the textual data of customer complaints of this bank in order to identify the influencing factors in reducing customer satisfaction and improving service quality. For this purpose, a combined method for entity extraction - based on thematic text modeling - and sentiment analysis has been introduced simultaneously. In this work, different methods based on deep learning have been evaluated. Experimental results indicate that Bi-GRU Capsule model has performed much better and has 88% for sentiment analysis and 60% for topic modeling for 57 very close classes.