Year: 2025
Type: article
Abstract: This article aims to start the development of language models for agronomic advisory service. This is done by evaluating various n-gram language models with different smoothers: Modified Kneser-Ney, Add-k, and Absolute. The models were constructed using an earlier collected dataset on Norwegian agriculture. They were adapted to provide practical agronomic advice on integrated pest management. Model performance was measured using perplexity. The Add-k (k=0.1) scored perplexities of 1920, 7800, 13600, and 16200 for 2-, 3-, 4-, and 5-grams, respectively. In this study, the Modified Kneser-Ney (D=0.8, D=0.8, and D=0.8) performed best, achieving perplexities of 494, 363, 344 and 339 for the same orders. When expressing the best performing model, the Modified Kneser-Ney model (trigram) could predict sentences such as “Meadowgrass is a grass that grows in more or less dense lawns”. However, before being practically useful, the models need further development.
Author: Kristian Nikolai Jæger Hansen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: bronze
OpenAlex ID: W4414929132
FWCI:
Cites:
Cited by:
Related to: