The Research On Writing Style Modeling Method Based On Supervised Learning

With the rapid development of mobile Internet technology, the more activity rely on computer people engaged in, the more need for intelligent computer can understand and process vast amounts of natural language information. In natural language processing, study and calculation the speech feature of personal and literary works is one classic research of Computational Linguistics, which is is difficult to obtain research breakthroughs.Writing style is a subjective description, and there is no a rigorous mathematical model capable of expressing it. The idea of the research is to study definition of writing style, define the writing style of formal mathematical model and test it by machine learning, then modify the theory by experiment, and get a more rigorous mathematical expression.Starting from this research ideas, firstly, consult the literature of Computational Stylistics, combined with previous work and writing style theory, using the mathematical symbols to define the form of wrting style, and described how to use a machine learning algorithm to learing writing style.After defining the writing style model, the paper brings forward a whole overall modeling programs, and explains the key issues in modeling process in detail.In order to verify the validity of the model and modeling, the paper carrying out the feature of writing style extraction, characterization and writing style recognition. Firstly, in the feature of writing style extraction and characterization, the paper proposed a clustering analysis to automatically extract style features, then use statistical methods to characterize the interquartile range of writing styles. And the final experimental results verify the writing style definition. Secondly, in the writing style recognition, the paper proposed Support Vector Machine algorithm used to create the writing style classification, and experimental results show that automatic extraction of features optimized can enhance the recognition accuracy. Lastly, the research brings forward thow two kinds of writing style ensemble learning algorithm to improve Support Vector Machine classifier, and the experimental results proved these algorithms is work.