Differential evolution based mention detection for anaphora resolution

Mention detection is an important component in anaphora resolution. In this paper we present our works on mention detection based on differential evolution (DE). The proposed technique consists of two steps, viz. feature selection and classifier ensemble. In the first step the algorithm performs automatic feature selection for two machine learning algorithms, namely Conditional Random Field (CRF) and Support Vector Machine (SVM). The first step yields a population of solutions, each of which represents a particular feature combination. We generate several models from these feature representations, and combine their decisions by a DE based ensemble technique in the second step of our algorithm. Experiments with a resource poor language show the recall, precisiommeasure valueseasure valuesn and F-measure values of 67.33%, 88.60% and 76.51%, respectively.