Recognition Of Uyghur Musical Named Entity Based On CRF

Music is neither restricted by ethnicity nor by border,plays a crucial role in daily entertainment.Uyghur music,in particular,has a long history and the Uyghurs are well known for being innately talented both at songing and dancing,having a uniquely fervent passion for music.Following the ever continuing development of the internet,a wide range of music-related information is now easily accessible to all.It has hence become necessary to make use of a series of functional software to make searching more easy and undertake the targeted treatment of musical information.More importantly,the pressure from the recent software development has given rise to the advent of research into a range of software functions,such as musical searches,musical editing,personalized recommendations and popular music trends.Therefore,conducting a research to develop techniques for both the Named Entity Recognition(NER)of Uyghur music and the treatment of information related to Uyghur music is not only of vital importance,but also has strong academic significance.This paper takes Conditional Random Fields(CRF)as its basis of research,with aim to conducting the Named Entity Recognition of Uyghur Music.The particular entities that shall require recognition include Artist Name,Song Title,Band Name and Album Name.Currently,the CRF model,being one of the most widely used sequence labelling models in the field of natural language processing,is a well-accepted model for conditional probability.It not only conquers the independence assumption of generative models,but also avoids the bias towards graphic model labelling.For this reason,this thesis not only considers sequence labelling as the most appropriate model for the Named Entry Recognition of Uyghur music,but also utilizes CRF models to complete the required tasks.In order to undertake Named Entry Recognition,the first task one must do is collecting appropriate data.This paper focuses its data collection on Uyghur websites,such as Karwan,Alkuyi and Hawar Tori,with the aim of organising,labelling and categorising the collected materials,created dictionary of related music entites,pre-treated materials and a series of work.Due to the fact that no linguistic material from the field of Uyghur music has yet to be labelled,the processing this particular section became lengthy and somewhat tiresome.Thereafter,according to the patterns found in the linguistic materials,particular features,such as context,dictionary and keywords are selected.A detailed description of the process of selecting features and establishing feature models are then described.Finally,the study not only assigns different window sizes to different features,but also designs experiments into the successive overlaying features,and makes comparisons between other models and the model used herein.Such experiments prove that CRF models are not only both feasible and effective for use in the field of Uyghur music,but also have a clear advantage in terms of accuracy when compared with other models.