Research And Implementation Of Subjective Question Scoring System Based On Chinese Word Segmentation And Text Similarity

In recent years,with the development of artificial intelligence technology,more and more people have begun to pay attention to the research of intelligent scoring.The current automatic scoring system can accurately score objective questions such as true or false questions,multiple-choice questions,and fill-in-the-blank questions with fixed answers.However,for subjective questions,it is currently mainly based on manual review and is being reviewed.Subjective test papers may be affected by personal emotions,page layout,and physical fatigue.The objectivity and work efficiency of the test papers are evaluated.Therefore,the study of subjective test scoring systems is of great significance to intelligent education.This paper proposes a research on a subjective question scoring system based on Chinese word segmentation and text similarity.The main work is as follows:1.Aiming at the existing Chinese word segmentation models based on neural networks,which usually require a large number of labeled sentences for model training and other problems,this paper proposes a Chinese word segmentation model based on dictionary information to make full use of the useful information of the dictionary and reduce Dependence on labeled data.Through the design of pseudo-labeled data generation and multi-task learning methods,the dictionary information is added to the neural network CNN-Bi GRU-CRF model for training,and the experimental verification is performed on the two benchmark data sets PKU and MSRA,and the accuracy rates are up to 97.6% and 97.8%,the experimental results verify that the method can effectively improve the performance of Chinese word segmentation.2.This paper proposes an improved multi-model text comprehensive similarity algorithm.First,a hybrid algorithm based on knowledge and corpus is used to calculate the term similarity of the text,and the similarity of short texts is calculated by introducing a dynamic VSM model and term similarity;second,a syntactic similarity algorithm based on the constituency parse tree CPT is proposed.To obtain the syntactic similarity of the text.Aiming at the sentence similarity algorithm,a sentence similarity algorithm based on non-linear weighted TF-IDF algorithm,improved Jaccard similarity coefficient and Word2vec-CNN is proposed.The algorithm as a whole obtains the similarity of the final text from the comprehensive consideration of the semantic similarity of the short text,the similarity of the syntax and the similarity of the sentence.Experimental results show that the accuracy rate reaches 83.0%,the recall rate is 90.0%,and the F1 value reaches 86.4%.Compared with the algorithm that only considers sentence similarity and text semantic similarity,the accuracy of the algorithm is increased by 15.1% and 5.7%,respectively..3.This paper designs an automatic scoring system based on subjective questions,and realizes the functions of each module.The system is trained and tested on a political question paper of a junior high school in Shanxi Province,and compared with the results of manual evaluation.The experiment proves that the system is in subjective questions.It has good performance in scoring,and the difference with manual scoring can be controlled in a relatively reasonable interval.