Emotional Computing For Social Media

With the continuous development of computer and network technology,the Internet has now entered the era of social media.The new online media represented by Weibo contains a large number of emotional texts on topics such as news,current affairs,policies and regulations,and consumer products,reflecting the opinions,sentiments,attitudes and inclinations of individual users.Researches on sentiment itself and the interaction between sentiment and other cognitive processes have attracted widespread attention from scholars,and sentiment computing has also become an emerging field of research.Emotion analysis is a basic task of sentiment computing,designed to automatically identify the emotions contained in the text,such as like,surprise,anger,and so on.This paper mainly studies multi-label emotion classification and emotion cause recognition in social media emotion analysis tasks.The first two chapters of this paper introduce the current research status and basic technology of this kind of problem.Then,in view of the shortcomings of the existing research,solutions to the task of multi-label emotion classification and emotion cause recognition for social media are proposed in the third chapter and the fourth chapter respectively.(1)To deal with multi-label emotion classification for social media,a joint binary neural network algorithm is proposed.In our algorithm,the text representation obtained through a neural network is no longer sent to a softmax function,but is sent to a set of logistic regression functions.Therefore,multiple binary tasks can be completed synchronously in the same neural network.In addition,we designed a joint binary cross entropy loss function to capture the relationship between different labels.Considering the characteristics of multi-label emotion classification task,we further propose to incorporate a priori emotion information into the loss function.Experiments show that the proposed algorithm outperforms the existing multi-label emotion classification algorithm in classification performance.(2)To deal with emotion cause recognition for social media,a neural network model fusing relative position information and global label information is proposed.We introduce a relative position augmented embedding learning algorithm,and transform the task from an independent prediction problem to a reordered prediction problem.We concatenate dynamic global label information in the hidden layer of the neural network to constrain the prediction of different clauses in the same sentence.Experimental results on a benchmark emotion cause dataset show that our model achieves new state-of-the-art performance.Further analysis shows the effectiveness of the relative position augmented embedding learning algorithm and the reordered prediction mechanism with dynamic global labels.