Based On Chinese Semantic Role Labeling Feature Selection Research

In recent years, with the mature of the lexical analysis and ayntactic analysis in the naturallanguage processing, it is the inevitable choice that semantic analysis promotes the further of thenatual language processing. Semantic analysis has become an important research direction innatual language processing. Semantic role labeling is a kind of shallow semantic analysis ofnatural language. The basic unit of semantic role labeling can be syntactic elements, phrases,words or dependencies. Now the research of semantic role labeling focuses on a pharse structuresyntax or dependencies. Because semantic role labeling based on phrase structure parsing proneto produce corpus sparse issues, semantic role labeling based on dependency parsing arousedextensive attention from researchers. Therefore,this thesis uses the dependency as mark unit tocarry out the study of Chinese semantic role labeling based on feature selection.Fristly, this thesis conducted an investigation about of the research status on the semanticrole labeling. It was carried out detailed exploration and analysis on domestic role annotation bybibliometric, include a summary of theoretical and applied research, research hotspots and trendsanalysis.Secondly, the applications of semantic role labeling technology in the information retrievalmodel were explored. Aimming at for the problem of limited natural language in the semanticretrienal model, semantic role labeling technology for improving the performance of informationretrieval model was analysised.Finally, we studied the Chinese semantic role labeling based on feature selection. Based onthe interdependent relationship characteristics, this paper analysised the issue of Chinese SRLcharacteristics, and optimized Chinese dependency characterisitics from the syntax andsemantics. Then we made a small-scale training experiment for Chinese SRL based ondependency syntax analysis, and selected effective features for improved Chinese semantic rolelabeling performance.