Automatic Coding Method Of Evaluation Questions In Questionnaire Survey And Its Application

In the questionnaire survey,in addition to most of the multiple-choice questions,some open-ended questions are also designed in order to have a more comprehensive and accurate understanding of the real thoughts or opinions of the respondents.However,due to technical constraints,currently the collected text data of this type of open-ended question is mainly carried out on the basis of manual coding,which has problems such as long time,high cost,and difficulty in standardization of results.Based on this,this paper uses topic classification model,sentiment analysis and other Chinese natural language processing methods to study the automatic coding method of evaluation open question text data,which lays a good foundation for further statistical analysis of questionnaire surveys in the later period.First of all,based on the analysis of relevant literature,we sorted out and summarized the main characteristics of the evaluation open questions in the questionnaire survey,and found that the questions often include "what","how","thought","why",and "can it?" and other question words,and the collected text data has the characteristics of short text,strong professionalism,strong subjective color,and many degree adverbs.Based on these characteristics,it is proposed to encode Chinese text data in open questions from the two aspects of topic classification and sentiment analysis.Secondly,for all the text data collected for a certain evaluation type open question,on the basis of preprocessing such as text segmentation and removal of stop words,the LDA topic model is used to obtain the corresponding topics and the keyword items of each topic,and then formulated the coding rules for the corresponding numbers of the topics;then use the TF-IDF algorithm to obtain the vector representation of each evaluation text data,calculate the similarity between it and each topic,and classify the topics according to the principle of maximum similarity,and then complete coding of the evaluation text data.Then,for each topic in the evaluation text,on the basis of constructing an emotional dictionary,by setting the weights of emotional words,negative words and degree adverbs,a corresponding quantitative model is established to obtain the emotional tendency of each topic in each evaluation text It divides the emotional tendency into five polarities,namely strong positive,positive,neutral,negative,and strong negative,and establishes an emotional coding method for the corresponding letters of the subject's emotional tendency.Finally,the topic classification coding is combined with the emotion coding of the corresponding topic,and the coding method of the evaluation type open question of "topic classification + five-pole emotion" is constructed,which can realize the automatic coding of single topic and multiple topics of the evaluation text data.A certain degree of innovation.The experimental results show that the F value of subject classification and sentiment analysis for each evaluation text data can be stabilized at about 70%,and the automatic coding model and method have high effectiveness.