Chatbot para pedidos online em restaurantes baseado em identificação de intenções e extração de entidades utilizando a ferramenta Rasa Open Source

Abstract

Due to the social isolation measures imposed with the advance of the COVID-19 pandemic, many establishments – including restaurants and snack bars, were forced to process orders online for delivery and pick up, to keep the operation on. One portion of these requests are made through messaging applications, where there is direct interaction via text between the customer and the establishment. However, this interaction can be sometimes problematic: for the customer there may be a negative experience due to delays and order ambiguities. And on the establishment side, there is the difficulty of scaling this service as the number of orders grows, the decrease in the productivity of employees who have to perform online service beyond their other duties, in addition to the order loss. In order to solve some of these problems, this project have applied natural language processing techniques – using the Rasa Open Source framework, to create a chatbot that is able to interact and understand the intentions of customers, automatic processing orders via chat in restaurants and snack bars. A chatbot prototype was developed and evaluated based on the performance metrics of the classification tasks – from data collected from conversations in a simulated experiment, resulting in 94% accuracy for the intention identification task.