AgReview - Hybrid Dataset for NLP-based Price and Sentiment Evaluation in Agri Inputs

The primary objective of this collection is to support AI-driven methods for cotton crop disease control and product recommendations. The two key components of dataset are comprehensive product listings and user-generated ratings, and it was selected from BigHaat, one of India's leading digital agri-platforms. The product data includes the product name, company, pricing (MRP, selling price, discounts), and packaging size (in milliliters, grams, LTR, etc.). Each item is categorized as Low, Medium, or High using quantile binning and standard deviation-based methods to allow for price intelligence while accounting for size-normalized pricing. This helps identify trends in market affordability and enables fair comparisons of different product volumes.
The reviews part, which documents real farmer comments, includes review texts and contextual agronomic information, such as dosage, usage instructions, and target diseases or insects. According to Shaver's Emotion Model, NLP techniques were used to classify emotions (satisfied, neutral, sad, and disgust) and sentiments (positive, neutral, and negative) using the CardiffNLP Twitter-RoBERTa model. Examining these results by hand improved accuracy.
This dataset is designed to support intelligent systems that assist farmers and agri-tech platforms in making informed, cost-effective, and emotionally conscious decisions. It will also significantly advance research on AI-powered product recommendations, price optimization, and user experience in agriculture.