Result on the average precision, recall, and F-measure using our features and domain composition kernel in the best average F-measure case for each set of features.

As sets of features, (F1–5), (F1–6), (F1–5,7), and (F1–7) shown in Table 1 were used. The results by the naive Bayes-based method [11], MCL [7], MCODE [28], RRW [18], and NWE [19] are also shown, where the experiments for these methods were performed by [11]. (B1), (B2:CC),, (B6) indicate the features by [11] (shown also in Table 3).