Research And Implementation Of Subjective Automated Assessment System Based On Natural Language Processing

At present, a favorable universality and practicability of subjective automated assessment doesn’t appears in examination system due to the restriction of natural language processing and AIT(artificial intelligence technology). Mostly the existing system at home deals with the subjective items by teachers’ manual work. By this way, it enormous affects the rapid expansion and the reform of teaching methods of education and teaching online. Meanwhile teachers spend a large amount of time in paper inspection, so that they can’t throw themselves in teaching with all their hearts. So it is important to do research in subjective automated assessment technology and realize it.The thesis starts with the thoughts of teachers’ manual assessment analyzing the whole process of manual assessment, and realizes the computer automated assessment by mimicking the process of teachers’ manual assessment. According the problems of lower accuracy in existing subjective automated assessment, combined with the characteristics manual assessment, the key point of automated assessment is how to improve the similarity of words and sentences. According the problems and deficiencies of calculating the words similarities in the How Net, this thesis proposes a kind of improved method, brings in synonyms dictionary, enlarges lexicon’s coverage area, and improves the accuracy of calculating the words similarity. In calculating sentences similarities, according the problems of lower similarity value, bring in calculation model of similar words to deal with the problems in the method of sentence similarity calculation which has the multi-features. At the same time, this thesis proposes calculation method of composition relationship similarity and improves the calculation method of sentence similarity. The experience shows that the improved method can enhance similarity between two similar sentences. And this calculation method is suitable for subjective automated assessment. When calculating the scores of examinees, improve the calculation formula by increasing the adjustment coefficient that reduces the error rate and makes the calculation more accurate.Finally, this thesis takes a subjective automated assessment system into practice. The system improves the calculation method of words and sentence similarity. The testing data show that this system has a good result. The system can meet the design requirement, and has a higher accuracy.