With the rapid growth of medical literature, it is difficult for physicians to maintain up-to-date knowledge by reading biomedical literatures. An algorithm named MedRank can be used to recommend influential medications from literature by analyzing information network, based on the assumption that "a good treatment is likely to be found in a good medical article published in a good journal, written by good author(s)", recomending the most effective drugs for all types of disease patients. But the algorithm still has several problems:1) the diseases, as the inputs, are not independent; 2) the outputs are not specific drugs; 3) some other factors such as the publication time of the article are not considered; 4) there is no definition of "good" for the articles, journals and authors. An improved algorithm named HIC-MedRank was proposed by introducing H-index of authors, impact factor of journals and citation count of articles as criterion for defining good authors, journals and articles, and recommended antihypertensive agents for the patients suffered from Hypertension with Chronic Kidney Disease (CKD) by considering published time, support institutions, publishing type and some other factors of articles. The experimental results on Medline datasets show that the recommendation drugs of HIC-MedRank algorithm are more precise than those of MedRank, and are more recognized by attending physicians. The consistency rate is up to 80% by comparing with the JNC guidelines.

Focusing on the issues in review helpfulness prediction methods that training datasets are difficult to construct in supervised models and unsupervised methods do not take sentiment information in to account, an unsupervised model combining semantics and sentiment information was proposed. Firstly, opinion helpfulness score was calculated based on opinion support score of reviews and replies, and then review helpfulness score was calculated. In addition, a review summary method combining syntactic analysis and improved Latent Dirichlet Allocation (LDA) model was proposed to extract opinions for review helpfulness prediction, and two kinds of constraint conditions named must-link and cannot-link were constructed to guide topic learning based on the result of syntactic analysis, which can improve the accuracy of the model with ensuring the recall rate. The F1 value of the proposed model is 70% and the sorting accuracy is nearly 90% in the experimental data set, and the instance also shows that the proposed model has good explanatory ability.

In view of the existence of a lot of spam comments in microblog, a new method based on AdaBoost was proposed to identify spam comments. This method firstly extracted feature vectors which consisted of eight feature values to represent the comments, then trained several weak classifiers which were better than random prediction on these features via AdaBoost algorithm, and finally combined these weighted weak classifiers to build a strong classifier with a high precision. The experimental results on comment data sets extracted from the popular Sina microblogs indicate that the selected eight features are effective for the method, and it has a high recognition rate in the identification of spam comments in microblog.

In order to avoid the problem of blind discretization of traditional classification "discretize first learn second", a new method of associative classification based on lazy thought was proposed. It discretized the new training dataset gotten by determining the K-nearest neighbors of test instance firstly, and then mined associative rules form the discrete dataset and built a classifier for predicting the class label of test instance. At last, the results of contrastive experiments with CBA (Classification Based on Associations), CMAR (Classification based on Multiple Class-Association Rules) and CPAR (Classification based on Predictive Association Rules) carried out on seven commonly used quantitative datasets of UCI show that the classification accuracy of the proposed method can be increased by 0.66% to 1.65%, and verify the feasibility of this method.