Annotated (Pseudo-)Transitive Relations of the LOD Cloud

This deposit consists of the manually annotated gold standard, graph files, and evaluation results for our study of (pseudo-)transitive relations. The <strong>gold-standard.zip</strong> file provides the files consisting of manually evaluated triples. <br> The files were exported from ANNit with the column:<br> - LEFT* for the source URI of the triple.<br> - RIGHT* for the target URI of the triple.<br> - UserChioce for the choice of user when manually evaluated<br> - Decision* for the actual decision made by rater/annotator. It can only be unknown, remove, remain.<br> - Comment, if any. The corresponding gold-standard.pdf file provides all the details about the creation of the gold standard. The folder graph_file includes the unweighted graphs, as well as the two sets of weighted graphs: the graphs with counted weights and the graphs with inferred weights (in the subdirectory of <strong>counted_weights</strong> and <strong>inferred_weights</strong> subdirectory respectively). <br> The files are compressed in the format of *.gz. Each file consists of two columns of integers as the source and the target. The integers correspond to the URIs. The corresponding mapping files are in the directory <strong>mapping</strong>. The corresponding files (of unweighted graphs) in WebGraph format are provided. These files were used when evaluating our algorithm against the exiting web-scale feedback-arc-set algorithm. The source code for refinement is at https://github.com/shuaiwangvu/Refining-Transitive-Relations Some raw data for table 2 and table 3 in our corresponding paper and their analysis are also provided. There were two static settings of the parameter for Table 2 and we chose the first setting for the final presentation in the paper. XL (static): b1 = 15,000 and b2 = 3,000 L (static): b1 = 1,000 and b2 = 200 Should there be any problem with these datasets, please feel free to report to us at the following email address: [e-mail]. A link to the full paper will be published when the paper is accepted.