Reasearch And Application On Multi-element Information Credibility Model For Microblogging

The problem of measuring information credibility on social networks has receivedconsiderable attention in recent years. Rumors, spreading fast and affecting widely, cancause great adverse effects to social, business or individual. It is a great challenge to de-termine the authenticity of a message on timeliness and accuracy. Therefore, building in-telligent and effective model, measuring information credibility is very urgent and impor-tant. Based on Natural Language Processing, Machine Learning and Behavioural AnalysisTheory, this thesis has completed modeling. The main work of this thesis manifests in thefollowing two aspects:Firstly, research of identify water army on micorblogging. Users played leader role onmicorblogging, so identifying water army is an important process in gauging informationcredibility. Most of existing methods of identifying water army is manually, it is difficultto achieve good performance in big data environment. This thesis thoroughly comparedand analyzed various characterizes and behaviors of normal users and water army, con-structed effective features on idenitying water army, based on probabilistic graphical mod-el proposed WGM model. Two datasets from Sina and Twitter microblogging system wereselected and comprehensive comparative experiments were conducted. Experimental re-sults show that features and WGM model is very excellent in identifying water army, andWGM model is more accurate and effective than traditional machine learning methods.Secondly, research of detecting rumor and gauging information credibility. Gauginginformation credibility, discriminating information authenticity is very important and dif-ficult. Based on comments in micorblogging, this thesis constructed three features measurethe difference of comments in rumors and veritable microblogs. Combining these features,this thesis proposed BPCM model to evaluate information credibility and detect rumors. Adataset collected from Sina microblogging system is used to evaluate performance of thesefeatures and model. Experimental results show that using comments can effectively evalu-ate information credibility, the features and BPCM model constructed in this thesis is veryexcellent and more accurate and effective than traditional machine learning methods..The models proposed in this thesis contribute to national public opinion monitoring,business marketing analysis, events analysis and detection.