| These scoring techniques of computer automated scoring system for objective questions, for example, multiple-choice questions, true or false items, fill in the blanks and other kinds of questions, has been very mature, These Systems have been applied to large-scale examination system. However, the automatic scoring techniques of subjective questions, such as, explanation of terms, short answer question, essay questions and other kinds subjective questions, due to the limited by the theory and technology development of artificial intelligence, natural language understanding, pattern recognition, has far from perfect, there are still no practical systems. The speed of computer processing is fast, high-efficiency, high precision, and never get tired, as well as its work is not affected by external factors, especially suitable for the automatic assessment of subjective questions, and teachers freed from heavy marking examination papers. Therefore, it has a great practical significance to study how to use the computer to achieve the automatic correcting of the subjective question.This article addressed two aspects of the automatic scoring of the subjective question; they are natural language processing technology and the domain ontology. The main works are shown as follows:(1) Building a domain ontology of data structure, the ontology provides a shared basis for communication between man and the computer, also facilitate the understanding between the system developers and researchers in various fields. In the module of parameter extraction of question, if the extracted parameters match with a concept in the ontology, it limited the scope of the problem answer and played a foreshadowing role in building the standard answer framework. In the sentence framework extraction of the standard (students) answer module, when enter a search terms, and then apply the ontology reasoning, then extent its semantic, and retrieve the associated words, then compare the keywords of student answers. In the pronouns digestion module, ontology library function is similar to the function of the module of parameter extraction of question. By the introduction of ontology, the data of ontology are given clear data semantics and structural description. It is understood by the computer, and achieved data interoperability in the semantic layer. In order to calculate the similarity between the words for better, this paper establish two ontologies, they are "data structure" and "programming language"(2) Processing of subjective questions, at present the automatic scoring system that have existed and being studied almost no handing the subjective questions. The automatic scoring systems for subjective questions examine the knowledge points and allocate the weight of sentence framework of standard answers according to the parameters from the questions. Analysis of questions is a very important part of automatic question answering system, the article also with the help of questions analysis to understand the subjective questions, and converting to the way that computers can understand extract keywords. After only by correctly understanding the meaning of the expression of subjective questions, we can identify the points of the questions, and then find out the score points of the standard answers and assign a higher weight.(3) Pronoun resolution standard (student) answers, the phrase of "refer to" is the semantic association between the current demonstrative pronouns and the phrase (antecedent) that appear above sentence. Pronoun resolution is the process of the current indicator to determine the antecedent. Even the answers to subjective questions, the applications of the pronoun, such as it, this, that, the former, is also very common. There are some these pronouns in the sentence of standard (student) answers, if you don’t digestion and direct assembly into the sentence framework. So the meaning of the pronoun is not very clear, it is not clear to refer to which objects. If the pronouns appear in the standard (student) answers, this can not be the comparison between sentence frameworks, It may be wrong even they can.(4) Building the standard (student) answer sentence frameworks, in the standard (student) answer sentence frameworks, some sentences are used to explain, while other sentences is the key points of scoring, therefore, the weight distribution is very important. The key frameworks is bound to a big weight, some sentences that used to explain or irrelevant to the connecting role can be assigned less weight. In the5.5sections, this paper has given the algorithm of building sentence frameworks and distributing the weight to sentence frameworks. After analyzing syntactic and dependency relationship for a sentence, then this paper extract event and entity relation to composition of a sentence framework. In the past of automatic scoring system for subjective questions, they are only considering the similarity of key words or sentences between students’answers and the standard answer and do not considering the weight of the sentences. Contrasting to the structured sentence frameworks between the standard answer and students’ answer, this doesn’t cut off the sentence semantic links.(5) Word similarity calculation based on domain ontology, to determine the similarity of a student answer and the standard answer, we can first calculate the similarity of the standard sentence framework and the students’answer. This paper obtains word similarity according to the already well-established domain ontology. On this basis, we receive student scores. In summary, this paper builds a new subjective Automatic Marking System, through extracting the parameters by processing subjective questions, building the domain ontology as a discipline system knowledge of examination questions, anaphora resolution for the standard answer and students’answers, construction of a sentence framework and take advantage of the vector space model to calculate the similarity of sentence frameworks. Compared with the past of the automatic scoring system for subjective questions, this system as much as possible to allow the computer to understand "semantics" as the core, to analyze the intent of the teachers want to check study points, to maintain the semantics of the standard answer and student answer contact, rating accuracy has improved. |
PDF Full Text Request