The Credibility Degree Measure Of Web Content In Social Media

The credibility of network information is related to the degree of effectiveness of decision making.In recent years,the new media represented by the Internet has gained rapid development and popularity,social media,such as social networks(SNS),micro-blog,We Chat,blog(Blog),with its unique features: user generated content(User-generated content,UGC),has become a major channel for Internet information release.This brings a variety of "title party","network rumors","false news" and other false news,making the public on the authenticity of the content of the network information to some degree of doubt.At present,the literature about the credibility of social media network information mainly focuses on exploratory research on the empirical analysis of influencing factors.The influence factors such as information specialty,accuracy,fairness and transparency will be directly or indirectly affect the information content of credibility;in addition,compared to the verbal information,the information content of the quality and credibility of online reviews with visual information obviously will obtain higher evaluation.Although the existing research results have promoted the research and development of information content credibility assessment,there are still some limitations.Therefore,this research mainly involves two parts:(1)the real problem: social media the popularity of user generated content formed by bottom-up public participation,the network information content messy and fragmented,exacerbated the generation and dissemination of false information network.Face the real measure of massive dynamic information difficult problem,method measure network information credibility;(2)scientific problems: the existing lack of quantitative qualitative evaluation measure,according to the information of user behavior data objectivity based on its cognitive experience and evaluation of social tagging of information content generated by credibility measure objective to construct a quantitative modelIn the social media environment,this paper considers the participants scale,condition attribute number,reference object number of the three factors,respectively,based on Bayesian inference theory,Bayesian network theory and transfer learning theory,constructing network information credibility measure model of the scene,the main research contents include the following aspects:(1)This paper constructed the credibility measure model based on Bayesian inference theory,and constructed the minimum error rate evaluation model for credibility degree measure based on Bayesian decision theory.With the increase of participants in social media,the minimum error rate of credibility degree will go down,and the credibility measure model based on Bayesian theory will be better than it based on fuzzy theory.The influence factors of the reliability measure error rate just focus on the scale of the participants.The other factors,such as the conditional attributes and the reference objects,should be considered in the future.This paper revealed the trend of minimum error rate with the increase of participants based on collective wisdom theory.(2)Under the assumption that participants scale invariant under the premise of considering the influencing factors of condition attribute range.Web contents are mainly text information,and its content can be embodied in keywords.Based on the theory of Bayesian Network which can reveal the conditional relationships between keywords,and the crowd wisdom which is emerged in the UGC of social media,this paper constructed the model of relative credibility degree measure.For the evaluation of the model,this paper constructed the minimum error rate evaluation model and the classification accuracy model.By three real data sets in different domains,the experimental results show that with the increase of condition attribute keywords in the condition of certain participants in social media,the contextual relative credibility degree of target attribute keyword will tend to decrease and the accuracy will tend to rise.(3)Under the assumption that the participants' scale and scope of condition attributes is fixed under the premise of considering the influence of the number of reference object.Based on the theory of transfer learning and the conditional correlation between different reference objects,credibility measure model construction of network information content,the experimental results revealed that with increased number of reference object,the ROC curve of the overall upward trend,and the area under the curve of AUC value change trend is more and more big.In view of the complexity of the factors involved in the measurement of the credibility of the network information content,there are still some work to be further improved in the future.