An instance based dynamic generalization model for coreference resolution

The traditional coreference resolution methods use immutable models to handle all kinds of instances, once the model created, all the parameters and weight are fixed, and thus acquire good results in common examples, but neglect long tail instances. To deal with this problem, we present an instance based dynamic generalization algorithm (IDGen). IDGen constructs an exclusive model level by level automatically for every single instance with the generation points created from the features. In every generalization level, we choose the best generalization point according to some criterions. The experiment shows that our IDGen algorithm is superior to the traditional methods.