Abstract: Aiming at difficulty of agricultural keyword extraction in domain text, an agricultural keyword extraction method was proposed, which combined new word discovery and improved TextRank algorithm．The algorithm calculated word formation probability of words in text through information entropy to find domain proper nouns and new words, and expanded word segmentation dictionary through manual audit．Based on word segmentation dictionary, calculation method of TextRank algorithm node value in the construction of word graph was improved, word position and part of speech weight were added, and comprehensive weight of words was used to extract text keywords．Through experimental comparison, F value of this algorithm was 7.5% higher than traditional TF-IDF algorithm on average, and 9.8% higher than TextRank algorithm on average．The algorithm had certain practicability．