Enhanced Named Entity Transliteration Model Using Machine Learning Algorithm

Abstract

Named entities are important elements in Cross Language Information Retrieval (CLIR) system for locating relevant documents. For translating named entities, compression word format model was designed in the literature, particularly on person names. This model only maps the given source name with the target name stored in the database but it does not generates the target name. The same model is applicable only for the source names which have equivalents in the database. To overcome these problems, a new model called mapping and generation with dynamic learning is proposed. This model transliterates the named entities, particularly person names, places and organization names. The process in the proposed model is to compress the given source name and compare with the target database names. The model accurately maps the source name with the target database names, if the source name has a right equivalent target name and retrieves the equivalent and relevant target names. Otherwise it transliterates the source name and retrieves the relevant target names from the web. These new words need to be updated in the database automatically. To automate the updation process, a dynamic learning algorithm is designed and deployed. The accuracy of the proposed system is improved using learning algorithm when compared with manual updation process.