Research On Automatic Personality Detection Method Based On Multi-task Learning

In recent years,with the development of information technology such as big data and the improvement of computing platforms such as ubiquitous perceptual data and graphics processor,the development of artificial intelligence technology has been promoted rapidly.Artificial intelligence is a branch of computer science.It attempts to produce an intelligent machine that can think and respond like people.Personality is the combination of a person's emotion,behavior,motivation and thinking mode.It shows that personal preferences affect personal decisions.Therefore,the automatic detection of personality can be regarded as the simulation of human consciousness,thinking and other information.It is of great significance in social text mining,and has gradually become an important sub topic in the field of artificial intelligence emotion computing.Automatic personality detection refers to mining the potential personality traits of people from their social texts with the help of natural language processing related technologies in the field of artificial intelligence.The most popular method is to extract feature keywords with the help of external domain dictionaries and classify pedestrians with the help of downstream machine learning methods.In view of the rapid growth of online social texts,feature-based methods require a lot of manual annotation and poor portability.Therefore,the automatic personality detection method based on deep learning has a wider application prospect.However,the existing personality testing model based on deep learning still has two limitations.First,they only extract the keywords corresponding to the potential personality,but lack the analysis of the emotional information and psychological characteristics of the poster.Secondly,due to the large number of online social texts,resulting in a long text sequence,some methods can not model the ultra long context,resulting in the neglect of many key semantic information in the post.In order to solve the above problems,a new multi label personality detection model combining emotional and semantic features is proposed in this paper.This paper mainly carries out the following research work:1.We propose a multi task framework combining emotion classification and personality detection,which automatically labels the documents through the external emotion analysis model,and uses the correlation between emotion and personality to improve the effect of personality detection.The framework is based on longformer pre training language model and fully captures the contextual language features of ultra long text data with the help of self attention mechanism.2.On this basis,we use the results of automatic emotion tagging to enhance the document data.By sorting the emotional degree of different posts,longformer can better capture the emotional association between posts,and further improve the effect of multi task detection.This model can automatically detect the personality of social media texts and help various social software to mine user information.In this paper,our model is trained on two public MBTI and Big Five data sets,and F1 is used as the evaluation index.The experimental results show that the model is superior to the most advanced technology in personality testing.Affective analysis model.