Research On Rumor Detection Mechanism In Micro-network

The widely use of micro-blog,WeChat and other social platform shorten the information spreading period and expand the scope of information spreading.So the influence and harm caused by rumors becomes larger.It becomes a hot issue to research how to detect and block rumors in the field of information spreading.In this paper,we detect rumors based on Maximum Entropy Model,the improved Maximum Entropy Model and explosion degree of rumors.In general,this paper mainly completed the following work:(1)The maximum entropy model is used to detect rumors and we determine the feature function and design training set for experiments according to characters of rumor.We find the optimum number of features for rumor detection.We compared rumor detect results with the results of SVM,BP-neural network model,Bayesian Model and K-means algorithm,and the accuracy of rumor detection method based on Maximum Entropy Model is equivalent Bayesian Model and K-meams algorithm.Thus,there is still room for improvement.(2)In this paper,I proposed a new method for constructing sample: the central distance clipping method,which is used to solve the problem of boundary blur and isolated points in the classification of imbalanced data.This method uses weighted vectors to represent each piece of information and uses the distance between vectors to represent the similarity among different information.I use the distance from each piece of sample information to the center of each kind of information to define isolated point and cut samples near boundary.The method solves the problem that the original sample has a blur boundary and many isolated points.I propose a new feature selection method: Difference Count method.This method gives full consideration to the influence of number of features on rumor detection.It also fully considers that the reference value of features which appears many times in both of rumors and non-rumors is lower.On the basis of this consideration,we calculate the difference count value of each feature: fDC)(,and sort the features according to the difference count value.Then we select N features with the maximum value of fDC)(.At the same time,I improved the feature function of the maximum entropy model to fit rumor detection.After constructing the mechanism of the rumor detection based on the maximum entropy model,I do the experiments of the rumor detection.In the experimental design,the selection of the training set is improved,and the central distance method is used to optimize it.I find the optimal number of features in micro-network.I also compare the experiment results of improved maximum entropy model with results of original maximum entropy model and the results of SVM,BP-neural network,Bayesian model and K-means algorithm.The experimental results show that the detection accuracy rate after optimizing the training set and feature functions is better than before.The detection accuracy rate is better than other related classification methods.(3)For information that is difficult to detect using rumor detection method based on Maximum Entropy Model,we detect them base the explosion degree of rumor.This paper build a game model between rumor makers and disseminators and on-Trust)ET(Explosi model.This paper find the common characteristics of the widely spread rumors through experiments based on the two models.The explosion degree of widely spread rumors is between 0.695 and 0.795.Thus,the explosion degree is an important basis of rumor detection.