Automatic Scoring Technology Of Subjective Questions Based On Textrank+word2vec And Its System Design

The computer-aided teaching has already greatly appeared in the course teaching,but now the general computer marking technology is aimed at objective multiple-choice questions.So far,there is no widely accepted method and applicable technology for subjective question-and-answer questions.Due to the variety of subjective questions and the wide range of subject fields,subjective evaluation is difficult and has become a hot topic in the field of natural language processing(NLP).There is no automatic subjective question scoring system which can be widely used up to now because new method of subjective evaluation,many techniques need to be involved and difficulty in realizing it.In order to reduce the difficulty of the research,the subject corpus of natural science courses is chosen as the basis for the research.A prototype of the knowledge brief question scoring system based on "system Modeling and Simulation" course was implemented in this thesis.Firstly,Jieba participle and Wikipedia decommissioned words are used to process the text;secondly,the improved TextRank algorithm was used to extract the keywords.Secondly,Word2 vec was used to vectorize the extracted keywords,and a special vector model was constructed based on system Modeling and Simulation;thirdly,the result of calculating the cosine similarity of the vector was used as the basis for the evaluation score;fourthly,a course management system based on Django framework and mysql database to realize the management of test database and achievement were designed in this research.The improved keyword extraction algorithm was proved to have high accuracy by testing 100 test data obtained based on the test database.The result of the system score,compared with the result of the teacher score,was found that the difference between the designed scoring module and standard deviation of manual score floating around 0.3,and more than half of the 100 samples obtained have a score error of less than 0.5 points.It was proved that the automatic scoring in the range of 0.5 error can be accepted during the grading process.In this thesis,the system prototype was realized by using the theory of subjective question scoring,and the key technologies of TextRank and word2 vec are turned out to be feasible,Improved keyword extraction technique is effective.The prototype of the subjective problem scoring system is proved to have certain practicability and have reference value for future research.