Development and Validation of AI-Driven NLP Algorithms for Chatbots in Requirement Engineering

The present research focused on the use of artificial intelligence (AI) and natural language processing (NLP) techniques in the field of requirements engineering within software development. The primary challenge is the prevention of miscommunication between the customer and the development team. In the worst-case scenario, it might lead to the premature termination of the project. The aim of this project is to develop a prototype of a chatbot able to evaluate consumer needs and suggest potential requests. The first step comprised a thorough evaluation of the chatbot’s requirements, followed by the development of a prototype. Two transformer models have been developed to classify customer input, and an additional model has been established to generate suitable requests. The classification was obtained by assessing the level of detail of the provided user input using a classification model, as well as classifying them based on ISO 25010 (quality criteria for software). Both versions utilized the DistilBERT models as their foundation. A GPT-2 model was trained to generate the inquiry. This approach utilized ambiguous user inputs and generated inquiries to get further information. To determine the user’s intention, it was decided to use RASA software to train an intention module. This module will be able to differentiate between a user’s question and their intention to proceed with the acceptance procedure. The initial classification model achieved an accuracy of 0.7033, whereas the second model had an accuracy of 0.2784. Moreover, the output generated by the GPT model varies only to a limited degree. The quality of the model is directly influenced by the quality of the training data. Increasing the number of data points and balancing the classes can help enhance the model quality. Nevertheless, this scientific work presents a fundamental basis for the possible utilization of transformer models in the field of requirement engineering. Further exploration of the application of NLP approaches using transformer models to understand customer requirements has the potential to reduce the failure rate of software development projects.