The adaptive recommendation models in the agriculture information service platform are very important, which provides personalized recommendation information to farmers. Aiming at the special agricultural knowledge bases, by using the recommended methods based on content filtering, we have established the farmer interest model and the document feature model. In the two models, taking account of the influence of distribution of features in the different table space and the effect of HTML document structure on the feature weights, we have improved the accuracy of recommendation model by improving the traditional feature extraction algorithm. The experimental results show that with the increasing number of users of agricultural information recommendation models, the precision and recall rate of them are also increasing, the accuracy of them are also rising.