Utterance Intent Recognition for Online Retail

In Indonesian business environments, vital information required for operations is often available from centralized sources or through direct connection with relevant departments. Despite accessible information channels, stakeholders such as customers, business owners, and other parties frequently contact organizations directly to request confirmation or additional information. The volume of questions varies with the company's size, providing difficulty for larger enterprises. This buildup of questions frequently causes delays in responding to information demands. This research focuses on constructing chatbots developed with using chat feature conversation obtained from an e-commerce platform. As a result, the goal of this research is to build a task-oriented chatbot model tailored to enterprises, with the goal of efficiently addressing user queries within the chatbot's domain. The intent classification task is trained with a Gate Recurrent Unit (GRU) neural network architecture utilizing the Adam optimizer. This study investigates models for categorizing agent and visitor intentions for chatbot interaction between customer and business owner, demonstrating a 65% accuracy in agent intent and a 74% accuracy in visitor intent. The study highlighted the overfitting challenge in intent classification, implying that improving minority intent categories should be prioritized to avoid such concerns and improve model quality.