Table 4.

Average precision changes with CTD (training) + Triage (testing)

Chemical names	BASE	IXN	TOPIC	IXN + TOPIC
2-Acetylaminofluorene	0.6776	0.6776	0.6814	0.7096
Amsacrine	0.7202	0.7308	0.7468	0.7577
Aniline	0.7625	0.7542	0.7477	0.7677
Aspartame	0.4902	0.4958	0.5269	0.5388
Doxorubicin	0.8767	0.8828	0.8871	0.8937
Indomethacin	0.9608	0.9610	0.9621	0.9604
Quercetin	0.9186	0.9190	0.9162	0.9189
Raloxifene	0.7820	0.7803	0.7737	0.7661
Average performance	0.7736	0.7752	0.7802	0.7891

Again a leave-one-out train and test procedure is followed. The full dataset was downloaded from the CTD database and used to augment the training. Any duplicates appearing in both training and testing sets were removed from the training set. ‘BASE’ uses word features without substance/journal names. ‘IXN’ and ‘TOPIC’ mean semantic and topic features, respectively. ‘BASE’ features are used for all the experiments.