| With the continuous development of computer and natural language processing technology,the automatic English essay scoring system has become more and more mature,and has gradually been applied to college exercises and exams.The off-topic English essay analysis function is an important part of the automatic English essay scoring system,which is related to the robustness and credibility of the score.The existing automatic English essay scoring system does not have an effective off-topic analysis algorithm.However,most foreign related technologies are based on supervised algorithms,which require a large amount of corpus with the same essay to be tested.Therefore,there are significant limitations.In general,it is of great significance to study an unsupervised off-topic English essay analysis model.This thesis takes the Chinese essays of Chinese learners as the research object,and designs an off-topic English essay analysis model for the automatic essay scoring system.The model can identify off-topic essays without topic-specific training data,only essay and the topic are needed.It also can extract the sentences in the essay that are not related to the topic.In order to achieve this goal,the main research contents of this thesis are as follows:(1)Combining the distributed semantic space and the structured semantic space,a semantic representation model called hybrid semantic space is constructed.Through the hybrid semantic space,the degree of semantic association between English words and phrases can be accurately obtained.(2)A sentence-level English essay representation method is realized by combining the hybrid semantic space model and the improved smooth inverse frequency algorithm.(3)According to the actual situation of English writing in China,two types of completely off-topic essays are identified: the bad-faith essay and the essay irrelevant to the theme.Meanwhile,in combination with the English essay representation method,two algorithms are designed to realize the detection of the above two types of off-topic essays.(4)An English essay sentence extraction algorithm is designed for the English essays that are not completely off-topic,and then we design a topic relevance scoring method by calculating the proportion of the number of the off-topic sentences in the total number of essay sentences.(5)Finally,through the above steps,an off-topic English essay analysis model is constructed and then tested in the real data set.The results show that this model has higher accuracy and better practicability than the existing unsupervised off-topic analysis model. |
PDF Full Text Request