The Design And Implement Of The Rhetoric Recognition System Oriented Chinese Essay Review

In recent years, due to the number of Chinese relevant examinations increase rapidly as well as the people, the difficulty of the artificial scoring also increases, In that case, if we can score automatically and review it by using machines, it can save manpower to a certain extent, as well as improving the efficiency of scale. And if we can score as well as giving some comments to the students, which can make students improve writing skills specifically. Considering these aspects, in the future, the automatic Chinese essay review can make it a deeply integrated between the natural language processing and education, eventually, it will benefit the students.This research mainly introduces around the identification of the rhetoric related characteristics, using the methods of combining of the traditional methods and the deep learning methods. On the basis of the corresponding of the characteristics, we can give review results, finally, we will form a visualization system interface. Among them, the basis rhetoric related characteristics mainly include segment parallelism, period of parallelism, reference and graceful sentence.To the identification of the segment parallelism, we mainly use forward index and special storage structures, as well as some other algorithms.In the identification of the period of parallelism, we design some special algorithms to the identification of it. To the identification of the reference feature, we mainly use inverted index to do it. Also, we use N-gram model to deal with the corpus. To the identification of the graceful sentence including metaphor and personification, we use the deep learning method to carry on the design. On the details, we use neural network algorithm with artificial corpus to make a effective model of binary classification.In the identification of the rhetoric related features, we add the tolerance of the optical character recognition, And, we mainly use statistical methods, each feature must satisfy the precision and recall rate in more than 70% of the dual requirements.In the end, the system added the rhetoric related features overcomes the weakness in concentrated scoring to a certain extent, and narrow the gap to 30%~40% compared to the indicators by people and people. Also, according to the identification of the rhetoric related features, we can give related review results, which, including rhetoric related comments and underline the original sentence in the essay. Finally, we give a visual system interface to enhance the user experience.