Research On Automatic Scoring Of Subjective Questions Based On Natural Language Processing

In recent years,the development of internet technologies such as artificial intelligence and big data has promoted the education field to develop in a digital and intelligent direction.Educational methods such as multimedia teaching,online courses,and online examination systems have emerged.The current online examination system has matured the scoring technology for objective questions,but the scoring of subjective questions mainly relies on manual work.Realizing the automatic scoring of subjective questions can improve teaching efficiency and has great academic and application value.To address the current problem of missing semantic and syntactic structure information in the current automatic scoring of subjective questions,Relevant techniques in Chinese information processing were used in order to improve the subjective scoring system and make the calculation of scores more accurate.This paper uses techniques related to natural language processing to automatically score subjective questions,improving several aspects of the automatic scoring process to make the calculation of scores more in line with people’s cognition.In addition,in order to reduce the error between manual and automatic scoring,an adjustment factor is added to the formula for grade calculation.The main work in this paper is as follows:(1)Chinese word separation technology and related algorithms was studied,and a Chinese word separation algorithm that can identify professional words was proposed.The Chinese word separation algorithm uses the NLPIR Chinese word separation system and introduces the forward maximum matching algorithm to identify the professional vocabulary.The similarity of keyword vectors is calculated using TF-IDF and cosine similarity.(2)The word similarity calculation methods of How Net and Tongyici Cilin was combined,a dynamic weighting strategy was used to make full use of the information on the hierarchical structure and semantics of the words in the two different knowledge bases.The range of calculable words is expanded to make the calculation of word similarity more comprehensive and accurate.(3)With the help of LTP-Cloud,a Chinese sentence similarity calculation based on dependency syntax and word semantics was proposed,and then takes into account the influence of several semantic features on the degree of sentence similarity.The similarity calculation was performed on the triad of dependencies in two sentences,and different weights were assigned to different dependencies,which makes the sentence similarity calculation more comprehensive and more accurate in measuring the similarity of the meanings expressed in two sentences.(4)The subjective question scoring module of the automatic scoring system was designed and implemented.By using word separation algorithm and similarity algorithm on students’ answers and reference answers,keyword scores and sentence semantic scores were obtained,and then bring them into the score calculation formula according to the weight of each part to score students’ answers comprehensively.The system test data shows that the application of this paper’s method in the automatic scoring system of subjective questions is effective and feasible,which can effectively improve the teaching efficiency and help the students to get accurate scores.