Research On Automatic Marking Of Subjective Questions Based On The Computation Of Text Similarity

In the online examination systems, automatic marking technique of objective test questions has been well nature. That of subjective questions, however, such as term identifications, short-answer questions and essays is still inconclusive and unsatisfactory, which is restricted by the underdevelopment of artificial intelligence and nature language understanding. Consequently, it is of great practical significance to study how to achieve automatic marking by computers.At present, the majority of existed automatic marking algorithms are carried out by evaluating the similarity of the student's answer and standard answer textually. Unfortunately, the complexity of Chinese semantics, together with the existence of synonyms, makes severe influence on the effectiveness of the algorithms. This paper presents a hierarchical method, which regards the full text as a collection of sections, regards the section as a collection of sentences, and regards the sentence as a collection of words. A Hownet-based approach for words is employed to measure the similarity of words, and extended to the sentences, sections and the full text by generating the similarity matrix. For the politics discipline, the majority of subjective questions are designed to inspect the students' mastery of related knowledge points, and thus each answer generally centers on one or a few knowledge points. As a result, the knowledge points are presented first, and expounded with a detailed discussion to complete the answer. According to the characteristic, this paper tentatively put forwards an algorithm of matching the knowledge points based on the knowledge structure tree. It is empirically verified that our approach, which combines the above two algorithms, makes remarkable improvements on the veracity of automatic marking of politics subjective questions.