| Comparison is a common expression which often implies subjective attitude and emotion.With the development and popularization of Internet,more and more comparative sentences are contained in the Web.It is of great value to extract elements from comparative sentences.Due to domestic and foreign related research of comparative sentences is still in its infancy,many methods of comparative sentences identification and elements extraction can be explored.Taking comparative sentences as research object,this paper aims to explore an effective comparative elements extraction method.The specific work includes the following four aspects:(1)Implemented a Bootstrapping method for comparative elements extraction based on part-of-speech tagging information,which can automatically generate a large number of elements results from very little manual labeled data;(2)Implemented a comparative sentences identification model based on support vector machines,and explored the influence of comparative words,POS,chunk,distributed word vector and elements feature on identifying comparative sentences.The experiments show that chunk feature is of high recall while distributed word vector feature is of high precision.(3)Implemented a comparative elements extraction model based on conditional random fields.A series of features are firstly explored,then active learning are used to improve the scale of corpus,eventually the method got better performance of elements extraction;(4)Combined with comparative sentences identification model and elements extraction model,the paper implemented a comparative sentence elements extraction system. |
