Multi-dimensional And Multi-granular Event Sentiment Computing In Social Media

The advent of Web 2.0 era spawned a variety of new Social Medias,such as Twitter,Sina Microblog,and We Chat.Social Media has its special characteristics,like openness,interactiveness,real-time,dynamic and so on.All of these characteristics attract people to express or share their opinions and interests with others without the limitation of time,location,range of activities and relationships.The various types of emotions contained in the hot event can be spread and evolved in the real world and virtual world through people's interactive behaviors,as comment,forward and praise.Analyzing the multidimensional emotions of event can mine different emotion patterns,so as to measure the specific event emotion accurately.Analyzing the multi-granular emotions of event can know the event emotion from different granularity.It can not only make the event management be more targeted treatment,but also have a comprehensive and detail understanding of the different levels of event emotion.Thus,it can provide more targeted and accurate guidance for the decision making.However,the texts of Social Media event have feathers of massive-sparse,dynamic-heterogeneous,obscure-vague,which increase the difficulty of event emotion computing.Most of the research methods for current text emotion analysis in Social Media only measure the polarities of the text from positive,negative and neutral,which loses the emotional states and emotion patterns.Moreover,these methods do not analyze multi-granular emotions of the event,but only analyze the emotion of the texts by machine learning methods or are based on sentiment lexicon.It lacks the integrity for event emotion analysis.In order to accurately measure the emotions of Social Media event texts,we propose a multi-dimensional and multi-granular event emotion calculation method in Social Media.The specific research content is divided into four parts:(1)In order to solve the problem of mass and sparse of hot event texts in Social Media,we construct two models named the Profile and Emotion Computing Model(PECM)and the Dictionary Supervised Profile and Emotion Computing Model(DSPECM)respectively.The text words and labels are used as the input of the models,and the profile distribution and emotion distribution of the texts,the words distribution of the profiles and emotions are output by the models.Meanwhile,the words with definite emotion is used as the constraint condition of the model to enhance the accuracy of text emotion calculation.(2)In order to calculate the words emotion of the dynamic and heterogeneous text data of one event profile,we propose a word emotion computing method based on the Word Emotion Association Network(WEAN).Firstly,the WEAN is constructed to express the emotion of the text by using the words and labels from labeled texts,and the words emotion value is calculated through a finite iteration of the network.Secondly,we make use of the word emotion in the basic word emotion dictionary to modify the network and then recomputes the word emotion,which effectively overcomes the problem of emotion uncertainty of the traditional methods.(3)In order to identify the interaction between obscure and vague emotions,a text emotion model based on Emotion Resonated and Restrained Circle(ERRC)is proposed.On the one hand,we consider the mutual influence of emotions.ERRC is built to ensure that the main emotions of the text are kept while the secondary and noise emotions are weakened.On the other hand,we consider the words in different contexts have different emotional expression.To mine the specific emotional expression of the words,we propose the concept of emotion pattern.It can discover emotion patterns from the texts of event profile,so as to modify the emotion of the texts.(4)In order to accurately compute the overall emotion in different profiles of the event,we propose an event profile emotion calculation method based on the user importance and the text importance.Three levels(user level,fans level,texts level)are considered to compute user importance,and the type of interaction and the number of interactions are measured as the factors of text importance.Finally,the user importance,the text importance and the text emotion are taken into account to compute the event profile emotion of different dimensions.At last,we finish to compute the multi-dimensional and multi-granular event emotion of Social Media event.The research of this paper can be used to monitor public opinion,analyze emotions of different dimensions of hot events and accurately compute the main emotions of events,which can help people to quickly understand the changes of emotions in the process of event development and help decision makers formulate programs.It also can be used for product testing,analysis for the product's emotional state,to provide guidance for enterprises to improve products and services,to find the advantages and disadvantages of competitive products and to provide users with product purchase recommendations.