Towards domain independent named entity recognition

Abstract

Named entity recognition is a preprocessing tool to many natural language processing tasks, such as text summarization, speech translation, and document categorization. Many systems for named entity recognition have been developed over the past years with substantial success save for the problem of being domain specific and making it difficult to use the different systems across domains. This work attempts to surmount the problem by proposing the use of domain independent features with a maximum entropy model and a multiobjective genetic algorithm (MOGA) to select the best features. The methods used in this work are backed up by experiments of which the classifications are evaluated using two diverse domains. Conclusions are finally drawn and the outlook for future work is considered.