A proposta de um chatbot para e-commerce

Abstract

Chatbots are modern tools to meet the need to employ automation in communication between e-commerce (electronic commerce) and the user. Chatbots are robots capable of maximizing sales, consulting new markets through customer satisfaction surveys, solving customer service problems and answering frequently asked questions. In addition, chatbots can reduce costs with teams that make contact with the customer while increasing the need for specialized labor. The growing demand for efficient customer service, along with the proliferation of social media platforms, should drive the chatbot market in the coming years"(GRAND VIEW RESEARCH, 2020). The objective of this work is to propose, develop and apply a chatbot as auxiliary tool in an e-commerce and later analyze the behavior and infer how concisely the dialogue was conducted. The central idea is also to create a virtual guide on a sales site. In this guide, the user can ask questions about general doubts about the operation of the platform, website products and other peculiarities that involve commerce. This work used frameworks and development kits, such as Dialoflow, Glitch and WhatsBot. To assess the behavior and assertiveness of the proposed chatbot prototype, a survey of expected questions and answers, that is, with the intention of creating a confusion matrix capable of mapping the errors and successes in the charbot responses and thus having an evaluation metric. The true positive rate (correct questions or sensitivity) of the proposed final model was 93.55%. It was numerically inferred through the evaluation metrics that the chatbot improves its specificity rate (rate of false positives, or unexpected responses) as we add more training phrases to each registered intent. For future works, it can be suggested the implementation of an expansion of the virtual guide also for sales. In addition to bringing more profits to the company under study, a sales chatboot is capable of generating more data for qualitative analysis. In this work, in addition to the analysis of assertiveness through quantitative methods (errors and successes of the questions asked to the chatbot), the satisfaction of the user and owners was studied. In these areas, according to the users themselves, the proposed model also proved to be effective, friendly and promising.