| In recent years,conversational applied language technology based on deep learning has developed vigorously,and clinical psychological counseling is a typical field in which this technology is used to achieve goals.At present,there are many problems in the domestic psychological consultation industry,such as the imbalance between supply and demand,the uneven professionalism of employees,the imperfect service system,the low acceptance of the public,and the large consumption of time and financial resources.The urgent need for mental health makes people look at cutting-edge technology.Based on the above background,the psychological consultation robot is built by using the technology related to deep learning,and the key technologies such as the word vector algorithm and natural language generation model involved in the development process of the virtual service robot are improved and optimized,which improves the effectiveness of the model in extracting the feature of the word vector and can respond quickly and accurately according to the user’s intention.Applying the psychological consultation robot to the software system can effectively improve the quality of online psychological consultation services,enhance the interactive experience with users,alleviate the problem of insufficient supply side of the consulting industry in China,and meet the needs of the public for psychological health.The main research contents of this thesis are as follows:(1)Construct a corpus in the field of psychological counseling and conduct data preprocessing.Using the open source framework of Scrapy,natural language data are collected in a classified way from psychological consultation websites such as "One Psychology" and "Good Doctor Online".In order to improve the quality of the corpus,the crawled data is fused with the efaqa-corpus-zh open source corpus for data sources,and the obtained data is cleaned and extracted,and then saved as a file in the.conv format to complete the construction of the psychological consultation corpus.(2)Research on word vector algorithm.This paper proposes to add depression weighting factors to the SBERT model to form an improved SBERT model,which can better understand natural language and effectively extract the features of the corpus.After that,K-means clustering algorithm is used.According to the experimental results,this model has better clustering effect compared with other contrast models.The corpus is divided into three parts:family,school and emotion for subsequent experimental training.(3)Research natural language generation algorithms.This article draws on the idea of sequence confrontation networks and uses the BH-Seq GAN model as a training model for natural language generation.The generator is an Encoder-Decoder structure,and its skeleton is designed based on the Bi-GRU network.The discriminator uses a hybrid structure of high-speed neural networks and attention mechanisms,which improves the relevance and logic between generated statements.Using this training model,comparative experiments were conducted on the open source corpus and the supplemented corpus,respectively,proving the necessity of supplementing the open source corpus.(4)Construct the software system of psychological consultation robot.Encapsulate the psychological consultation robot model interface through the flask lightweight framework,use HTML to layout and display the front page,and realize the front and back end interaction.Through multiple rounds of dialogue on different topics,the system completes the performance and function test,and finally realizes the visualization function of automatic question answering and query.From the perspective of psychotherapy,this paper aims to establish a software system that can provide personalized psychological counseling services to meet the needs of different groups.From the theoretical level and the specific work flow,the feasibility and effectiveness of the psychological counseling robot software system studied in this paper has been confirmed. |
PDF Full Text Request