Research On Happiness Analysis Based On Joint Model

Happiness refers to the pleasurable emotions produced subjectively by humans.It is a positive part of emotions and affects the quality of people’s lives.Therefore,understanding human happiness is a worthwhile undertaking.This paper mainly explores the Agency and Sociality of happiness.In order to broaden people’s perception of happiness,this paper conducts research on happiness in English data set Happy DB and Chinese data set Weibo DB.In order to analyze the Agency and Sociality of happiness,a classification model based on transfer learning is proposed in Happy DB data set,and a joint model is proposed in Weibo DB data set.When Happiness is studied on the Happy DB data set,there are problems of short text length and small data set size.To solve this problem,this paper defines the Happiness analysis task as a text classification task with short sentence granularity and proposes a short text classification model based on transfer learning.In this model,we firstly retrain the pre-trained language model BERT with unlabeled training sets to obtain happy BERT,a semantic enhancement model for the target domain.Then,the language models BERT and happy BERT are further combined with other deep learning models to obtain a variety of single-text classification models.Finally,an improved voting strategy is proposed for model integration of multiple single models.Compared with the three baselines,the experimental results show that the proposed method is better than the baselines.When it comes to predicting the Agency of happiness,the proposed model gets an Accuracy of 0.8574 and an F1 value of 0.9000.When it comes to predicting Sociality,the proposed model gets an Accuracy value of 0.9280 and an F1 value of 0.9360.Happiness analysis is a novel task,and there is no corresponding Chinese corpus.In order to expand the research of Chinese happiness analysis,this paper constructs a Chinese happiness data set Weibo DB.At the same time,a joint model is proposed to study happiness in the Weibo DB data set.This joint model is composed of BERT model and an improved graph convolutional neural network.BERT model can capture the continuous of semantic and grammatical information,improve the figure of convolution neural network to the traditional figure of convolution neural network parameters is introduced in the figure,pooling method can capture the structure of the complex information from the perspective of figure and the global features,the joint model can take into account of the continuity of the text as well as the overall characteristics of the text.The proposed model is compared with six baselines,and the experimental results show that the proposed method is better than the baselines.When it comes to predicting the Agency of happiness,the proposed model gets an accuracy of 0.7523 and an F1 value of 0.7058;when it predicts Sociality,the proposed model gets an accuracy value of 0.8198 and an F1 value of 0.8021.