Skip to Main Content
Table 15 

Results for bridging antecedent selection: Comparing pairwise models with different feature sets and antecedent candidate selection strategies. “Ante.” and “Acc.” stand for “Antecedent” and “Accuracy,” respectively. The bolded score indicates a significant improvement over all other models (p < 0.01).

ModelFeaturesAnte. candidate listSettingAcc.
baseline1_NB Poesio features 2-sentence-window mention-entity 18.9 
baseline2_NB Poesio features 5-sentence-window mention-entity 18.4 
baseline3_SVM Poesio features 2-sentence-window mention-entity 19.8 
  
local1_SVM Poesio salience features + PrepPattern features (f5 and f6 from Table 12) 2-sentence-window mention-entity 29.1 
  
local2_SVM Poesio salience features + all features from Table 12  2-sentence-window mention-entity 39.3 
  
local3_SVM Poesio salience features + all features from Table 12  d-scope-salience mention-entity 46.0 
ModelFeaturesAnte. candidate listSettingAcc.
baseline1_NB Poesio features 2-sentence-window mention-entity 18.9 
baseline2_NB Poesio features 5-sentence-window mention-entity 18.4 
baseline3_SVM Poesio features 2-sentence-window mention-entity 19.8 
  
local1_SVM Poesio salience features + PrepPattern features (f5 and f6 from Table 12) 2-sentence-window mention-entity 29.1 
  
local2_SVM Poesio salience features + all features from Table 12  2-sentence-window mention-entity 39.3 
  
local3_SVM Poesio salience features + all features from Table 12  d-scope-salience mention-entity 46.0 
Close Modal

or Create an Account

Close Modal
Close Modal