Type Prediction Of Objects In Linked Data

In the linked data,the type information of the object is very important for the mining and utilization of the linked data.However,the problem of missing object type is common in link data.How to determine the missing type of object becomes an urgent problem in the field of data science.Studying on the traditional type inference to find the missing type information type object by logical reasoning,but the noise in the link data concept layer and the instance level causes an error propagation in the type inference process,which results in the failure of the type inference.Type prediction method based on machine learning is to get the attention of researchers.Since there are a large number of text information and links in the link data,the paper proposes a type prediction method based text and a type prediction method based link relationship.In the type prediction method based on text,the paper propose two concepts:object graph and virtual document.The object graph model simplifies the multi pattern RDF graph into a single pattern,the information extraction and classification algorithms are more friendly to the prediction process.The type virtual document is extracted from the text in the object graph,which contains an indicative text message for the object type.The type of object is a virtual document composed of multiple sub documents.According to the indication of the type of each sub document,the paper proposes four different weight allocation strategies of sub document.In the type prediction method based on the link relationship,the paper based on the classical ICA algorithm in the collaborative classification method,proposes a new method for data attribute and object oriented attribute prediction,and compares the data attribute and object attributes for prediction of the type indicator.Finally,the experiments carried out on several datasets verify the effectiveness of the proposed method.Text based and link based prediction methods have achieved good results in terms of precision,recall and F value.The experimental results show that the proposed method can effectively identify the types of missing objects in large scale link data.