Constructing and Cleaning Identity Graphs in the LOD Cloud

Five tables and six figures of the paper. Table 1 shows evaluation of 200 owl:sameAs links, with each 40 links randomly chosen from a certain range of error degree. The percentages between parentheses are calculated without considering the links evaluated as “can’t tell”. Table 2 presents accuracy of the approach on the four manually evaluated samples, based on a threshold of 0.99. Links evaluated as “can’t tell” by the judges are discarded. Table 3 shows precision, recall and accuracy, based on two thresholds (0.99 and 0.4) for the Barack Obama identity set. Links evaluated as “can’t tell” by the judges are discarded.Table 4 is comparison of the original identity network closure, the closure (b) and (c), with the Gold Standard. Table 5 shows precision, recall and accuracy evaluation of the three closures. Figure 1 is the workflow of the identity network extraction, compaction and closure. Figure 2 depicts the distribution of identity set cardinality in Gim. The x-axis lists all 48,999,148 non-singleton identity sets. Figure 3 shows error degree distribution of all owl:sameAs statements in the LOD-a-lot. Figure 4 is comparison of the original identity network and its transitive closure, with the two newly constructed identity subgraphs.Figure 5 shows‘Barack Obama’ identity cluster. Figure 6 presents community structure of the‘Barack Obama’ identity cluster.