Research On Truth Discovery Method Based On Information Relation And Label Confidence Clustering

Truth discovery is one of the key technologies to solve the problem of information conflict in the process of multi-source information fusion.This work can find the most reliable information from the information provided by different sources,which played a vital role in the process of data aggregation such as database,healthcare,social sense and crowdsourcing.Especially in the research of crowdsourcing aggregation,the process of truth inference and truth discovery are essentially the same.However,due to the diversity of information structure and the unreliability of sources,people are often used to modeling probabilistic graphs for specific scenarios.The setting of the parameters of the probability graph model and the complexity of the calculations restrict the large-scale application of the truth discovery technology.The optimization-based truth discovery method and cluster-based crowdsourcing truth inference algorithm are not only concise in calculation,but also have strong universality.Therefore,it is of great application value to study the optimization-based truth discovery method and clustering-based crowdsourcing truth inference algorithm.The main research work of this thesis is as follows:(1)Research on the optimized truth discovery method based on information relation.The information provided by the source to the object is one of the key factors in the research of truth discovery,and the relation between these information has a certain influence on the effect of truth discovery.This thesis proposes an optimized truth discovery method based on information relation.First,the method processes some redundant information and abnormal information,and then constructs the support function between the information by considering the relation between the information,which be added to the framework of optimization-based truth discovery.The block coordinate descent method is used to solve the parameters,which can obtain the reliability of each source and the true information of each object.Experimental results show that the method is superior to its comparison algorithms in accuracy.(2)Research on crowdsourced truth reasoning method based on label credibility clustering.The truth discovery problem in crowdsourcing labeling is the truth inference on crowdsourcing.This method aims at the single selection problem in crowdsourcing labeling tasks.First,the weights of each crowdsourcing worker are generated by the optimization-based truth discovery method.The credibility of each label of each task,using each label class as the feature of the task,and the credibility of the label as the feature value,according to these feature values,the K-Means algorithm is used to cluster all tasks into K clusters,and the K is number of label classes.Therefore,each cluster corresponds to a label class,thereby obtaining the true label of each crowdsourcing task,that is,the truth.Compared with some classic crowdsourcing truth inference algorithms,the method has higher accuracy,not slower calculation speed and stronger universality.