User Groups’ Sentiment Analysis And Recommendation Algorithm Based On Social Affordances

The influence of user groups’relationship on user sentiment is a widespread phenomenon in the emotional communication of social networks.It is beneficial to study the emotional tendency of user groups from several nodes that affect emotional transmission and emotional infection.It can help regulate the healthy development of cyberspace.It can help reduce the large-scale aggregation of negative emotions and reduce the occurrence of social emergencies.Therefore,this thesis studies from the following aspects.Firstly,in view of the absence and neglect of user group relationship logic in the existing text sentiment tendency analysis.This thesis proposes the concept of social affordances of user groups.It can be used to study the emotional tendency of the user group,that is,connect-ability,the structural closeness of user groups,greet-ability,the influence of users in user groups,and emotion-ability,the interaction frequency of users in user groups.First of all,this thesis puts forward the measurement index of social affordances of user group,and analyzes the correlation with the emotional similarity of users in the user group.The results show that there is a positive correlation between the social affordances of user groups and the emotional similarity of users,and the positive effect of emotion-ability and greet-ability on the emotional similarity of users is stronger than that of connect-ability.Therefore,considering the interaction of users’ emotions,this thesis puts the three attributes of users’ social affordances into convolutional neural network for sentiment analysis,and achieves good results on yelp data set.F1 value is 0.89%higher than that of logistic regression model method,1.72%higher than that of support vector machine method,and 0.62%higher than that of convolution neural network method without social affordances.This thesis also selects a business on yelp website to make a fine-grained emotional analysis of its product review text,and obtains the user’s emotional strength in four aspects:food taste,service,environment and price,which can guide businesses to improve their products.This thesis selects 16 businesses on yelp website,and makes a comparative analysis of the degree of emotional divergence and the average emotional score,which can help users measure the quality of businesses and choose the right businesses from another perspective.Secondly,the social affordances of user groups is a subjective factor that affects the mutual emotional infection of users in the same group to a certain degree.With the rapid development of big data and artificial intelligence technology,personalized recommendation system for different groups is an objective factor affecting users’emotional infection.Therefore,this thesis improves the existing recommendation algorithm,and proposes a recommendation algorithm based on graph convolution network,which uses graph convolutional network to express the characteristics of user-product interaction(purchase and score),user social relationship(friend relationship and social affordances of user group),and product-product correlation.Due to the powerful representation ability of graph and the rich social relationship information of users,compared with PMF,SoRec,GC-MC and GCNCF,the recommendation algorithm proposed in this thesis achieves better results in MAE and RMSE indexes.The MAE is about 0.18 less than PMF method and RMSE is about 0.17 less than PMF.The concept of social affordances of user group proposed in this thesis can help to study the mechanism of emotional communication of user group from the nodes within the user group.Meanwhile,the rich social characteristics of user group described by social affordances can help to improve the recommendation system.