| The examination is an important part of both the daily teaching and the variousactivities of appraisal and election. Examination questions can be divided into twoclasses: the objective items and the subjective items. The technology for automaticscoring of objective items is quite mature nowadays. But for various types ofsubjective items, such as the short-answer, essay and the writing part, the automaticscoring techniques are quite difficult to be implemented, for the reason that theanswer for the subjective items is descripted by natural language, which causes theautomatic scoring process must involve the knowledge of natural language processing,artificial intelligence and pattern recognition etc. Actually, how to use the computer toautomatically score on the subjective items has always been a key issue to be solved,and is also a hot spot in natural language processing area.After investigate the existing automatic scoring systems home and abroad, theresearch on techniques for automatic scoring of subjective items pays its attention tothe semantic similarity comparison between the examinee answer and the referenceanswer. And then a new automatic scoring model based on the result of dependencyparsing is put forward. This model uses techniques in the natural language processingarea, such as word segmentation, POS tagging and grammar parsing, in order to makepretreatment of the text about the examinee answer and the reference answer. Firstlyfunction words and punctuations, which have nothing to do with the similaritycomputation, will be removed. Then the algorithm of word semantic similaritycalculation will be involved in. By establishing the word similarity matrix, we will getthe result of the semantic similarity between the examinee answer and the referenceanswer. And Finally the automatic scoring system will show us the score.In the end of this paper, we have realized the automatic scoring prototype systemwhich is based on the automatic scoring model. By setting the control experimentsthat are established on the traditional automatic scoring system based on results of theword segment, we finally prove the correctness and validity of this proposed scoringmodel. |
