Aplicação de redes neurais convolucionais para análise de sentimentos para a descoberta de conhecimento sobre o cliente

Abstract

Customers exchange information, opinions and feelings daily on various topics linked to the products and services offered by companies. However, the data generated in these testimonials need to be analyzed to extract useful knowledge for the company to better understand the customer, in order to provide better service. Artificial Intelligence (AI) has methods and techniques applicable to the analysis of feelings expressed by customers in free texts with low structuring. Thus, the application of AI to discover knowledge in databases can help in understanding the feelings expressed by the client. The objective of this research is to apply techniques of convolutional neural networks for the analysis and classification of sentiments in customer comments aiming at discovering customer knowledge in a retail company. In addition, we also sought to compare the results of the experiments with the results of the detractor indicators of the NPS (Net Promoter Score) of a retailer. Therefore, this exploratory and experimental research was made possible through the execution of experiments based on the stages of the Knowledge Discovery Databases (KDD). Convolutional neural network techniques were applied for customer knowledge discovery. The main results of the research indicate that many customer comments are related to certain aspects of the company's products and services, where the following stand out: card, limit, payment, increase and difficulty in attending to the channels made available by the company. As for the crossing of attributes related to the client's profile with the NPS results, it was possible to identify the comments and main arguments segregated by gender, age group, income level and place of residence of clients with NPS detractors. As a conclusion of the research, it is possible to state that the developed solution can provide knowledge discovery of the client from texts elaborated by them. In addition, it is also asserted that the developed solution can assist in customer management in retail companies with a large volume of customer data.