With the vigorous development of the Internet, the network information grows rapidly, so does the agricultural network information. Extracting hot words from massive information is of great significance for monitoring and analyzing agricultural public opinion. Up to now, there is some research on hot words extraction, but there are still many problems such as poor pertinence. Existing hot word extraction methods cannot meet the personalized needs of users in different industries in agriculture. Therefore, a method of automatically extracting hot words based on agricultural network information classification was proposed. Firstly, the texts were classified by using the multi-label classification algorithm and multiple corpuses were built according to the classification categories. Secondly, the hot word candidates for each category were extracted by using the method based on information entropy. Thirdly, the heat of each hot word candidate was calculated by using the method based on time variation. Finally, these candidates were sorted by heat degree, and hot words were got according to the sorting results. Totally 15354 texts from agricultural websites were extracted for the experiment, automatically obtaining the hot words in the specified time period. The experiment results showed that the accuracy was over 0.9. It proved that the proposed method can extract agricultural hot words with high quality and help different agricultural user groups find and analyze the hot spot information of the industry.