Abstract:
Since chatbots were first developed in the 1960s, a lot has changed, and recent developments in Machine Learning and Natural Language Processing technology have greatly enhanced both their usability and adoption. These sophisticated computer programmes can simulate human speech and can communicate with users verbally or in writing. In a number of industries, including e-commerce, healthcare, finance, and customer service, they have shown to be a significant tool. Chatbots are an essential tool for organisations in the digital age because they automate time-consuming jobs, offer 24/7 customer care, and boost user engagement. The growth of e-commerce enterprises, however, might be hampered by inadequate customer assistance and linguistic difficulties, particularly in nations where the majority of the populace speaks a language other than English. E-commerce is a quickly expanding sector in Pakistan, and more businesses are going online to access a larger market. However, the growth of e-commerce enterprises in the nation may be hampered by inadequate customer assistance and linguistic difficulties. The creation of a Roman Urdu chatbot for e-commerce websites is crucial in this situation. Roman Urdu is a widely spoken and understood language in Pakistan, and having a chatbot that can converse in it can improve user interaction and boost client loyalty. Roman Urdu is written in both Urdu and English, making it difficult to create a chatbot that can understand it. Though it is now possible to create chatbots that can comprehend and reply to Roman Urdu queries because to recent developments in deep learning and NLP technologies. In this project, we developed a Roman Urdu chatbot for e-commerce websites using a Feed Forward Neural Network model and Flask framework. The chatbot was trained on a large dataset of Roman Urdu queries and responses using TensorFlow, Keras, and other major Python libraries. We used transfer learning to fine-tune a pre-trained model for our task, as transfer learning has shown promising results in NLP tasks. The chatbot’s architecture was designed to handle queries and provide fast responses to users. The chatbot performed admirably, obtaining over 90% accuracy on our test set. In order to test and enhance the chatbot continuously, we also integrated it with e-commerce websites. HTML, CSS, and JavaScript were used to create the chatbot’s GUI design, which offers a simple interface for users to communicate with the chatbot. The design of the chatbot could be expanded in the future to include more capabilities like sentiment analysis, entity recognition, and summarization. These features can enhance the chatbot’s functionality and give users more value. The chatbot’s performance can be improved by collecting more data to train it further, as more data can help it better understand the intricacies of the language. Roman Urdu chatbots for e-commerce websites have the power to revolutionise customer service in Pakistan by addressing the linguistic difficulties that hinder e-commerce companies.