《深度自然语言处理》课程教学课件（Natural language processing with deep learning）11 coreference resolution

Evaluation>MUCThere is a connected edge between any two mentionsNumberofentitiesinpredictionNumber of entities ingoldmSpred [-|P(Spred )]/Sgold1-[P(S gold )IZPrecision:Re call7[S goid |-1-i=lB3theentityclusterreferedbymentionminpredictionthe entity clusterpredogoldmRe call - I pred.goldrefered by mentionPrecisionIgold.mI predm Im in gold交通大学

Evaluation  MUC There is a connected edge between any two mentions.  B3        n i pred i m i pred i pred i S S P S ecision 1 1 | | 1 | | | ( )| Pr Number of entities in prediction        n i gold i n i gold i gold i S S P S call 1 1 | | 1 | | | ( ) | Re Number of entities in gold | | | | Pr m m m gold pred gold ecision   the entity cluster refered by mention m in prediction | | | | Re m m m pred pred gold call   the entity cluster refered by mention m in gold

Evaluation>MUCThere is a connected edge between any two mentionsNumberofentitiesinpredictionNumber of entities in goldmSpred [-|P(Spred )]/SgoldI-[P(S god )IL-Precision:Re callZis goid I-1-i=lB3MtheentityclusterreferedbymentionminpredictiontheentityclusterpredogoldmRe call - I pred.goldrefered by mentionPrecisionIgoldmII predm !m in gold交通大学CEAFΦ(g*)Φ(g*)Re callPrecisionZRZEthe entity cluster in predictionthe entity cluster in prediction

Evaluation  MUC There is a connected edge between any two mentions.  B3  CEAF        n i pred i m i pred i pred i S S P S ecision 1 1 | | 1 | | | ( )| Pr Number of entities in prediction        n i gold i n i gold i gold i S S P S call 1 1 | | 1 | | | ( ) | Re Number of entities in gold | | | | Pr m m m gold pred gold ecision   the entity cluster refered by mention m in prediction | | | | Re m m m pred pred gold call   the entity cluster refered by mention m in gold    i Pi g ecision | | ( *) Pr    j R j g call | | ( *) Re the entity cluster in prediction the entity cluster in prediction

Outline1.Background2.Classical Models3.Progress4. Datasets & Tools5.Future Work

1. Background 2. Classical Models 3. Progress 4. Datasets & Tools 5. Future Work Outline

ClassificationRule-based交道大学

Classification Rule-based

ClassificationRule-basedMachine learning-based麦通大学

Classification Rule-based Machine learning-based