Research And Design Of The English Essay Off-topic Detection System For Chinese College Students

With the development of English Essay Automated Scoring System like JUKUU and Bingo English, an increasing number of colleges begin to use Automated Scoring System to help teaching of English writing. Off-topic detection function is indispensable to an Automated Scoring System, but there is no a set of complete Off-topic detection algorithm in China at present, and the foreign related technology is not open. The study of a universal Off-topic detection algorithm for the teaching of College English writing is very important. In this paper, based on the characteristics of Chinese college students’ English Essay writing, an Off-topic Detection System is designed for the teaching of College English writing.By comparing the effect of current Off-topic Detection technology, the author design and implement an Off-topic Detection System from the point of suiting for Chinese learners. To achieve this goal, we attempt to proceed from the following four aspects:Firstly, we research the characteristics of the college students’ English writing in China and categorize the Off-topic essays by discussing with experienced English teachers. Secondly, we collect different types of English essays, then invite two professional English teachers to evaluate whether an essay is off-topic and its off-topic type. After that, we calculate the consistency of the two teachers to select corpus in this paper. Thirdly, we research the technology of off-topic detection in China and overseas, we put forward and verify three off-topic detection algorithms based on excellent models, text clustering and topic sorting according to whether the essays under test have excellent models or essays to be tested. Fourthly, we analyze the demand for the Off-topic Detection System to determine the implementation scheme and carry on the overall design and implementation of the system.First, we researched the writing characteristics of English essays of the Chinese college students to divide the off-topic essays into five types of that we determine three types as the objects of study. Then, we put forward three off-topic detection algorithms based on excellent models, text clustering and title sorting according to different conditions. Compared to the study abroad, the effect of the three algorithms are improved significantly, especially the accuracy, the recall rate and the Fl measure of the algorithms based on excellent models reached 60.56%,67.19% and 63.71%. Finally, we designed and implemented an Off-topic Detection System suitable for college English writing teaching according to the demand of the Off-topic Detection System.