Abstract:
Potato is the second most important food crop in Kenya after maize, and has been used to address food security challenges with declining yield in cereals. The yield in potato farming has been noted to decline gradually in the last few years due to limited supply of quality seeds as well as limited access to expert advice. This research addressed this challenge by designing, developing, and evaluating an AI-based chatbot model built using IBM Watson Assistant. The chatbot was intended to provide personalized expert advice and connect potato farmers with certified seed producers in a scalable and accessible manner. The specific objectives of the study were to review literature on chatbot technologies to inform the design and development of an AI-based potato chatbot model, to design and develop an AI-based chatbot model tailored for potato farmers, and to evaluate the model's overall effectiveness in addressing the needs of potato farmers in Kenya. This research presented a solution that offered expert advice and a link to quality seed producers through the use of Watson Assistant, which is an artificial intelligence-based chatbot framework. First, an introduction to potato farming in Kenya was presented. A quick review and the evolution of chatbot models marked the end of the first chapter. Chatbot theory was discussed in the second chapter, with a primary focus on its classifications, the general architecture of chatbots, and the Watson Assistant architecture. The methodology used in the research was discussed, with a primary focus on the four significant steps taken, which were data collection, data preprocessing, system development, and the model testing, training, and evaluation phase. The next section then presented the performance indicators that were used to evaluate the chatbot model in detail. The results were then interpreted and discussed, giving rise to the last section where conclusions were drawn. The first conclusion drawn was that the potato farming chatbot model achieved a score of 97.7% in message coverage, a score of 78.4% in conversation containment, and was found to have an overall effectiveness of 88.05%. User acceptance was at 60%, while the adoption rate was at 80%. Finally, recommendations on future work were presented based on user feedback and input from chatbot modeling experts.