India is considered as a major agricultural powerhouse as this sector counts to about 15% of India’s Gross Domestic Product (GDP). However, Indian agriculture still lacks access to modern farming techniques, which results in the wastage of resources. One of the modern technologies used to bridge this technological gap is an user-friendly messaging application known as Chatbot. A chatbot trained with agriculture related data can assist the farmers in having knowledge about the modern farming practices without the hassle of waiting for an expert to help solve their queries.
In this context, a natural language based chatbot model is proposed here, named KisaanBot, as a remote assistant to the farmer, which can respond to their queries. Our assistant is trained by the RASA Open Source Framework and can be accessed with a Web URL via Dialogflow deployment. The proposed chatbot model identifies the intent and the entity from user utterances and it generates the responses from the trained database, which delivers it to the user. The knowledge base used in training is designed by collecting grouped data from Kisan Call Centre (KCC) and Indian Council of Agricultural Research (ICAR). This knowledge base comprises the frequently asked questions and the corresponding answers regarding agricultural practices of West Bengal concerned with four domains namely soil, weather, crops and fertigation. Hence, the dataset comprises three fields, such as intent (the goal of a user query), query (the question related to that intent) and the answer (response from the chatbot). The chatbot is trained with 9 subqueries for each query to develop the knowledge base having 9 queries for a single intent. This can resolve the issue which arises due to different ways of framing a particular question by the users. The proposed system is pretrained and tested in the 80:20 ratio with the dataset. The efficiency of the proposed model is assessed by the performance measures, such as precision, recall and f1-score. The overall accuracy thus achieved by the proposed model is 99.82% which shows an improvement over existing works.
Keywords: Chatbot, Agriculture, RASA, Dialogflow, Natural Language Processing (NLP), Accuracy
This is B.Tech Final Year project of Ridam Hazra and Arpan Kesh, batch of 2023 at the Department of Computer Science and Engineering (CSE) in the National Institute of Technology Durgapur, under the guidance of Dr. Parag Kumar Guhathakurta.