Research On The Automatic Assessment Of Subjective Examination Questions Based On Sentence-Similarity

In the computer examination system, the automatic assessment technique aimed at monomial choice subject, multiple-choice test and filled blank etc. has developed consummately, applied the large-scale examination system. But the automatic assessment of subjective technique about brief-answer test questions has been faultful because of the development restriction of the artificial intelligence and natural language comprehension so that there hasn't been a applied system yet. In China, the manual's assessment is the main technique to resolve the subject tests, but the result of manual's assessment is influenced by subject or object factors, such as the neatness of examination paper, the normative ness and uglily of the font, the teachers' emotion etc., these factors reduce the competitive equitableness and honesty among the students. The computer is not affected by subject factors for its fast speed, high efficiency and precision, which makes it suitable for automatic assessment of subjective. Hence, it has operation significance to research how to use the computer to achieve the automatic assessment of subjective. In this dissertation the sentence-similarity was used for automatic assessment of subjective. The definition and the model based on sentence-similarity achieved action and Chinese particularity is analyzed, at the same time machine the techniques, such as translation, automatic summarization and information retrieval etc. are used to study the sentence-similarity, a new model based on the multi-level sentence-similarity is proposed based on above works in this dissertation. On the basis of evaluating the similarity according to the sentences' morphology, semantic and acceptation, amalgamating these three levels, it is more accurate to evaluate the sentence-similarity. By specific experimental tests, the automatic assessment of subjective system satisfied the demand of the system design basically. Finally, the effectiveness of the approach is verified by the experiments.