AgriResponse: A Real-Time Agricultural Query-Response Generation System for Assisting Nationwide Farmers

Abstract

Advancements in information sciences can play a vital role in strengthening the nation’s sustainable agriculture goals. In this direction, we propose a framework for a text-based query-response generation system to cope with the demand for timely help to the nationwide Indian farmers. One of the major challenges in designing such systems is constructing a knowledge base that can answer plant-protection-related questions from a diverse population of farmers. To tackle this problem, the past eight years’ call-log records from the countrywide farmers’ helpline network are collected and processed to construct the required knowledge base. Additionally, three response-retrieval models with approximate matching and spatial-based searching functionality are developed to administer the user input questions and extract relevant answers from the base. To validate the performance of the proposed framework, a diversified question bank consisting of 755 queries covering 151 crops in India is compiled. Three metrics (Accuracy Percentage, Crop-weighted Performance Score, and Average Response-retrieval time) are considered for the models’ assessment. Experimental results show that AgriResponse is a practical framework in real-world applications, with the different retrieval models useful for different scenarios.

Keywords

• Artificial intelligence in agriculture
• farmers’ problems
• helpline center knowledge base
• query response generation
• question answering system