Research And Validation Of Automated Assessment Model Of Subjective Tests

In recent years, with the rapid development of information technology, the importance of automatic grading becomes rather clear. Automatic grading avoids the huge consumption of human and financial resources in traditional exams and saves social resources. Also, automatic grading avoids any score bias caused by graders' subjective factors, ensuring the fairness and justness of the exams. However, current Chinese automatic grading systems can score the objective items only, such as multiple-choice questions and true or false questions. Automatic grading on subjective items involves natural language processing, artificial intelligence, and so on. Because of the complexity and openningnature of Chinese, Chinese automatic grading on subjective items still needs further investigation.Sentence similarity evaluation is one of the basic and key elements of automatic grading of subjective tests. The effectiveness and accuracy of algorithm of sentence similarity measurementss plays an important role on automated assessment model. This thesis starts with sentence similarity evaluation. Firstly, we introduce conventional sentence similarity measurement approaches and summarize the advantages and disadvantages of each approach.We then propose an improved semantics-based sentence similarity measurement approach. This improved algorithm makes use of global semantic information, sentence structure information and sentence semantic information to compute similarity values and then judges the semantic direction of sentences. It effectively improves the accuracy of sentence similarity measurements.This thesis builds and realizes an automated assessment model based on the sentence similarity measurements. The main idea of this model is to simulate the mental process of teacher reviewing. The model uses mainly text preprocessing, sentence preprocessing and sentence similarity measurements to complete automated grading. Then, we achieve this model by using the PHP language in the wamp development environment and design experiments that sets the automated assessment model which is based on the traditional semantic sentence similarity measurement as model A, sets the automated assessment model which is based on this improved semantic sentence similarity measurement as model B, and uses 100 manual graded subjective papers as the experiment samples. The final experimental data showed: 1. The model proposed in this thesis is feasible. 2. The improved semantic sentence similarity measurement proposed in this thesis effectively raises the accuracy of sentence similarity computing.