The Sample Selection Model Based on Improved Autoencoder for the Online Questionnaire Investigation

摘要

This paper presents the sample selection model based on improved autoencoder to solve low response rate in the online questionnaire investigation industry. This model utilizes the improved autoencoder to extract the samples' features and uses the softmax classifier to predict the samples' loyalty. Furthermore, the autoencoder is improved with three steps: first, the number of middle hidden layer nodes is determined by Singular Value Decomposition (SVD); second, the loss function of the autoencoder is improved with the information gain ratio; finally, the concept of Random Denoising Autoencoder (RDA) is introduced to enhance the robustness of the model. Through the selection model, samples with high loyalty will be picked out to answer the questionnaire so that the response rate can be improved. Experiments are performed to determine the feasibility and effectiveness of the model. Compared with the BP neural networks, the prediction accuracy of our model is totally improved about 8.5% and the success rate of sending questionnaires is also improved about 15%.