Research Of Push Model Algorithm Based On Individual Prediction

It is importantly that letting farmers get more information to the development of agriculture, with the coming of information society as well as the fact that china is a big agricultural country. How to acquire the useful, credible, and interested information from the massive amounts of information becomes one of the current research focuses of information service industry. The pull-based and push-based technologies are the two main channels of getting information. But the push-based technology better suited to rural areas, with the current situation that the rural development in China is relatively backward and the cultural qualities of rural users are generally low.Building a high-precision prediction model of information pushing is the most critical part of the system, while "k-nearest neighbor selection" is the core issue of push model. This study improved the pushing model from two aspects of similarity measurement and k-nearest neighbors focus on "k-nearest neighbor selection"Neighborhood formation is the most important step of push model. The completeness of the neighbors affects the overall quality of push model. Neighborhood formation consisting of k samples of users with the highest similarity measure, commonly used similarity with Pearson correlation coefficient, cosine similarity and similarity of mean-square deviation. But these measures do not reflect the complex non-linear relationship between the two users, causes the neighborhood formation not great enough. In this paper, Maximum mutual information coefficient is introduced as similarity measure between samples, compared to the traditional mutual information, MIC respond effectively to complex non-linear relationship between the two users, making Neighborhood formation more great, improving the prediction accuracy of push model.Predicting not scoring project of target user, based on Neighborhood formation, is another key point of push model, prediction score value of the project directly decide whether to push the project to the target user, errors can lead to errors of prediction information push.Build project score prediction model with high resolution, choose the appropriate training sample is the key.Neighbor set is based on all have been scored for project calculation similarity, but in the prediction of a particular user’s specific project,because of time difference, regional difference, the existence of cultural differences and so on, with all the neighbors samples as the training sample may not get the best prediction effect.From all neighbor set select the k optimal samples is a k-neighbor selection problem, the choice of the k value is the core.Introduction to statistics, this study analyzed each forecasting project near to the collection of structural, given a common variation A, and from all the neighbors set for each user choose training sample who’s distance is less than k. realize the personalized prediction of each user.Based on above neighborhood formation and improvement of training sample selection in two parts, the support vector machine model of MovieLens DataSet has a greatly improves of the item rating prediction accuracy.