Research On Hyperparameter Optimization Method Of Emotional Computing Model Based On Machine Learning

Deep learning has strong learning ability,and can automatically extract useful feature representation from data by supervised learning algorithm.It has been widely used in image recognition,speech recognition,knowledge Graph,natural language processing and other fields with outstanding performance.Similarly,in the field of affective computing,the application of deep learning algorithm has achieved great success.In order to improve the shortcomings of traditional parameter adjustment methods,this paper proposes a automatic hyperparameter search method,and applies it to two important tasks in affect computing field: speech emotion recognition and depression automatic detection.The purpose is to solve the problem of difficult parameter adjustment in affective computing field and improve the recognition rate and prediction accuracy of these two tasks by finding the best parameter configuration.The main research work of this paper includes:Firstly,aiming at the problem that it is difficult to adjust the parameters of deep learning model proposed in the field of affective computing,combined with the advantages of visual analysis of regression tree in machine learning,an automatic hyperparameter search algorithm is proposed.Its main idea is to choose the appropriate algorithm model for the tasks in the field of affective computing,Secondly,the search space is set in stages for the hyperparameters in the model;Finally,the search process from coarse to fine is carried out in the search space,a set of hyperparameter configurations are randomly combined,the algorithm model is trained,the validation loss and test loss are obtained,and the results are input into the regression tree for analysis,focusing on the search space.Second,the proposed hyperparameter automatic search method is applied to speech emotion recognition task.The experimental data comes from the IEMOCAP data set,based on the TXT file provided by the data set to describe the emotion in speech,four kinds of emotions are selected from nine kinds of emotions to construct the data of this task.The experimental results in the emotion classification database constructed in this paper show that the automatic hyperparameters search can effectively find a set of optimal hyperparameter configurations and achieve good results.Thirdly,the proposed automatic hyperparameter search method is applied to depression automatic analysis task.The experimental data comes from AVEC2017 data set,and the experimental results show that the automatic search method based on hyperparameters can quickly and effectively obtain better prediction results.