In order to support the suitability of the PCA* Confidence metric for prediction of new facts, we carried an experiment which uses the rules without constants mined on YAGO2 (training KB) to predict facts in the newer YAGO2s (target KB). We took all rules mined by AMIE with head coverage threshold 0.01 and ranked them by standard and PCA confidence. Then we took every rule and generated facts beyond the KB using our experimental setup. The following table shows top 30 rules ranked by our PCA confidence and their actual precisions beyond the database. For further details, please refer to the detailed evaluation information as well as to the full list of rules mined by AMIE on YAGO2.
Rule | Precision in the unknown region | Std. Confidence | PCA Confidence | New Predictions | Total predictions |
---|---|---|---|---|---|
?b <isMarriedTo> ?a => ?a <isMarriedTo> ?b | 100.00% | 53.01% | 91.79% | 5158 | 5158 |
?a <diedIn> ?c ?c <isLocatedIn> ?b => ?a <isPoliticianOf> ?b | 13.33% | 2.70% | 85.38% | 13038 | 13038 |
?a <livesIn> ?c ?c <isLocatedIn> ?b => ?a <isPoliticianOf> ?b | 13.33% | 2.47% | 83.33% | 4939 | 5137 |
?a <isLocatedIn> ?c ?c <hasOfficialLanguage> ?b => ?a <hasOfficialLanguage> ?b | 70.83% | 0.07% | 83.33% | 75579 | 75579 |
?a <isMarriedTo> ?c ?c <livesIn> ?b => ?a <livesIn> ?b | 79.31% | 33.71% | 69.88% | 326 | 326 |
?c <isMarriedTo> ?a ?c <livesIn> ?b => ?a <livesIn> ?b | 60.71% | 22.81% | 68.42% | 299 | 542 |
?c <hasOfficialLanguage> ?b ?c <isLocatedIn> ?a => ?a <hasOfficialLanguage> ?b | 30.00% | 14.29% | 64.29% | 100 | 100 |
?a <created> ?b ?a <produced> ?b => ?a <directed> ?b | 0.00% | 49.83% | 58.68% | 976 | 976 |
?e <isMarriedTo> ?a ?e <hasChild> ?b => ?a <hasChild> ?b | 34.48% | 57.57% | 57.57% | 1865 | 1865 |
?a <wasBornIn> ?c ?c <isLocatedIn> ?b => ?a <isCitizenOf> ?b | 3.33% | 3.75% | 57.42% | 32491 | 32491 |
?a <wasBornIn> ?c ?c <isLocatedIn> ?b => ?a <isPoliticianOf> ?b | 0.00% | 0.65% | 57.18% | 30883 | 33559 |
?a <isMarriedTo> ?f ?f <hasChild> ?b => ?a <hasChild> ?b | 33.33% | 56.11% | 56.11% | 1068 | 1986 |
?c <isLocatedIn> ?b ?a <isLocatedIn> ?c => ?a <isCitizenOf> ?b | 3.45% | 0.40% | 54.24% | 52354 | 52367 |
?a <isCitizenOf> ?b => ?a <livesIn> ?b | 100.00% | 4.50% | 52.36% | 13913 | 13915 |
?a <influences> ?f ?f <isCitizenOf> ?b => ?a <isCitizenOf> ?b | 3.33% | 10.56% | 51.96% | 1173 | 1199 |
?a <livesIn> ?c ?c <isLocatedIn> ?b => ?a <isCitizenOf> ?b | 0.00% | 3.91% | 48.13% | 4167 | 5047 |
?a <diedIn> ?c ?c <isLocatedIn> ?b => ?a <isCitizenOf> ?b | 0.00% | 4.51% | 48.13% | 10681 | 12791 |
?a <actedIn> ?b ?a <created> ?b => ?a <directed> ?b | 3.45% | 38.29% | 45.23% | 518 | 608 |
?a <diedIn> ?b => ?a <livesIn> ?b | 84.00% | 1.21% | 43.41% | 21883 | 21889 |
?a <hasChild> ?c ?b <hasChild> ?c => ?a <isMarriedTo> ?b | 70.00% | 17.32% | 40.98% | 1170 | 1720 |
?a <diedIn> ?c ?b <hasCapital> ?c => ?a <isPoliticianOf> ?b | 20.69% | 0.16% | 39.60% | 22069 | 24784 |
?c <hasAcademicAdvisor> ?a ?c <graduatedFrom> ?b => ?a <worksAt> ?b | 3.33% | 13.44% | 39.04% | 552 | 552 |
?a <actedIn> ?b ?a <created> ?b => ?a <produced> ?b | 0.00% | 20.64% | 38.98% | 790 | 790 |
?a <diedIn> ?c ?c <isLocatedIn> ?b => ?a <livesIn> ?b | 100.00% | 1.31% | 38.55% | 12643 | 13225 |
?a <isCitizenOf> ?c ?c <hasCapital> ?b => ?a <diedIn> ?b | 0.00% | 5.05% | 38.46% | 12936 | 12936 |
?a <created> ?b ?a <directed> ?b => ?a <produced> ?b | 4.35% | 16.11% | 37.47% | 4973 | 5258 |
?a <livesIn> ?b => ?a <isCitizenOf> ?b | 3.33% | 4.49% | 36.57% | 12609 | 13611 |
?a <directed> ?b => ?a <created> ?b | 20.83% | 18.75% | 32.46% | 26980 | 26980 |
?a <livesIn> ?b => ?a <isLeaderOf> ?b | 3.33% | 1.47% | 31.39% | 14287 | 14287 |
?a <livesIn> ?b => ?a <diedIn> ?b | 100.00% | 1.83% | 29.05% | 13949 | 14140 |
*PCA=Partial Completeness Assumption