| In recent years, agriculture-related information technology has been developing continuously and acceleratingly, the usage of Internet in rural areas increases spirally and the application of Internet has reached a new level. The massive agriculture-related knowledge and information online brings the convenience for people, but it also increases the difficulty of information retrieval. According to the "Xinjiang rural information acquisition system" project, it studies the classification of web page information on the Xinjiang unique crop, and implements the effective classification method, with the purpose of obtaining obtain useful information accurately for agriculture-related staff members in Xinjiang.Four machine learning methods have been studied are Rocchio, kNN, SVM and Na?ve bayes, and through experiments, it combines the two ideas of clustering and classification algorithm.How to cut the training set of samples which is outliers is the focus of study.In this paper, the main work is as follows:(1)Ba Zhua Yu software provides agriculture web page crawling, then remarking all the samples with a classfication label, namely, policies and regulations is marked as 1, agricultural science and technology is marked as 2, cotton is marked as 3, corn is marked as 4, wheat is marked as 5, walnut is marked as 6, jujube is marked as 7, grape is marked as 8.(2)Some key technology before text classification was studied. Finally, we use the Pao DingJieNiu software to segment the page content, then take the feature extraction in CHI, 140 of the highest scores are selected in every categories. Four supervised learning classification algorithm ut Rocchio, kNN, SVM and Naive Bayes are used to deal with the prepared data and to classify agricultural information in experiment and analysis. Finally, according to the results, it showed that both SVM and kNN are good enough.(3)This paper proposed a combining of SVM and K-means method, and according to keeping center vector and holding adjacent samples of center vectors, we take the comparison of two methods in training time, and F1 measure score.Summary, this paper studies the classification of web page information on the Xinjiang unique crop.The train set is 11200, the test sets is 4800. Using Rocchio, kNN, SVM, Naive Bayes algorithm to classify the web pages respectively, and analyzing the classification results and performance of each algorithm: the SVM classification have better measure precision, recall and F1 score, however, in the case of large data, SVM takes more training time; K-means is very fast, but the accuracy is so poor. Therefore, the paper proposes a method of K-means combined with SVM, it puts forward two methods of reducing training data to achieve the purpose of reducing the training time. The results show that both of two methods can reduce the training samples and save training time. |
