Research And Implementation Of Algorithms Of Tibetan Micro-blog User Recommendation

This thesis is for the Research of personalized recommendation based on Tibetan User of Sina Micro-blog, since Tibetan informationization development lags behind because of various factors resulting in Tibetan microblog text personalized recommendation research are temporarily not.The paper grab the data through the Sina micro-blog API, and through denoising, encoding conversion, segmentation and part of speech tagging and filtering of stop words and so on, with the auxiliary help of manual and computer to build the micro-blog Tibetan corpus; secondly, extracting the feature on behalf of the micro-blog user by the unsupervised contribution method, and using TF-IDF weighting method, finally using the VSM to build personalized recommendation engine; then, considering the various recommendation algorithms and combined with the characteristics of micro-blog itself and the Tibetan research status, finally decided to choose the collaborative filtering based on Clustering Recommendation algorithm; finally, the experimental scheme of two kinds of algorithm is compared to the standard K-means and K_means++ that the result of the two algorithms by experiment, the standard k-means algorithm as a baseline, with the results are detailed analysis, based on the recommendation system performance evaluation refers to the standard accuracy, recall, and F1 values, thus draws the relevant conclusions of the experiment, and finally completed the personalized recommendation system.According to the final experimental results show that the improved K_means++ algorithm compares with the K-means clustering algorithm, the F1 value of the K_means++ algorithm is 0.052 higher than the K_means algorithm in the baseline, achieves the better performance and meets the demand in the plan.The innovation of the paper is the first personalized recommendation of Tibetan microblog users,followed by the contrast of the two clustering algorithm that K_means++ algorithm does better than K-means clustering algorithm, and finally, I achieve the recommendation system. In this paper, the further research will start with how to build a better recommendation engine, and choose a more effective algorithm to carry out the experiment.