Entity Sentiment Calculation And Inference Based On Emoji Sentiment

Nowadays,due to the openness and popularity of Internet,hot events spread quickly.Meanwhile the frequent and novelty of hot events also promote the prosperity of Internet.When hot events occurred in geographic space are reported by the Internet,social media space will generate a lot of discussion about hot events in a short time because of its advantages of real-time publication,strong interactive and easy to use.Therefore,a comprehensive overview of hot events in geographical space-Internet space-social media space is formed.Hereinto,social media(e.g.Weibo,Twitter,Facebook,etc.)is becoming a most active platform that provide space for a large number of users to discuss hot events.Everyone expresses opinions on hot events by publishing social texts,which contains rich sentiment.And it is very necessary to grasp users' sentiment in order to provide decision-making guidance for handling hot events.Therefore,sentiment analysis of hot events based on social texts has become a hot research field.However,the existing methods of sentiment analysis have the following limitations: 1)With the popularity of emoji(such as(?),(?)),more and more researchers have begun to use emoji for sentiment analysis.But so far,few sentiment analysis work have calculated emoji's sentiment quantitatively;2)Entity is the object of emotional expression.The current methods mainly calculate entities' sentiment through vocabularies.There is little work analyzing entities' sentiment through emoji and vocabulary patterns;3)The entities in hotspot event are not isolated,and there are few work calculating entities' sentiment by entity emotional reasoning;Aiming at the limitations mentioned above,this paper proposes three methods to solve them: 1)We achieved emojis' six-dimensional sentiment automatically and quantitatively through an improved label propagation algorithm—Label Attenuation Propagation Model.2)We construct three-layers network(entity layer network,vocabulary and vocabulary patterns layer network,and Emoji and Emoji mode layer network)firstly.Then,based on the three-layer network,the sentiment-enhancement-suppression algorithm is proposed to implement entity sentiment calculation.3)An entity equivalent relationship and entity-verb bipartite graph based method is proposed reason entity sentiment.The specific research content of this paper is as follows: 1.Based on the co-occurrence relationship between Emoji in social media,we build the Emoji Link Network to organize the massive and sparsely confusing Emoji in social media.Some Emojis in social media can be used in multiple contexts,which leads to the sentiment uncertainty of Emoji.So,we propose an emoji sentiment uncertainty measurement method based on the PAD three-dimensional sentiment model to measure the sentiment uncertainty of Emoji.Combining Emoji Link Network and the sentiment uncertainty of Emoji,we propose an improved Label Propagation Algorithm-Label Attenuation Propagation Model to calculate the six-dimensional sentiment(Love,Joy,Anger,Sad,Fear,Surprise)of Emoji automatically and quantitatively.Emoji with sentiment uncertainty will introduce a lot of noise to emoji-based social text sentiment analysis.In order to determine emoji's sentiment,we have tapped the emoji combo in social texts.So that we get emoji patterns which have more certain sentiment.Then,we calculated the sentiment of Emoji patterns.2.Based on the co-occurrence between emojis and vocabularies in social media texts,we use emojis and its patterns to express the sentiment of vocabularies and its patterns.Based on the the constituency parse result of the social text,we calculate the distance between the entities and the vocabularies,so as to use the vocabularies and its patterns to express the sentiment of entities.Based on the steps above,we build an entity-vocabularies and vocabulary patterns — emojis and emoji patterns three-tier networks to express the sentiment of entities richly.Based on this three-tier network,we propose an emotion-dynamic enhancement suppression algorithm to compute entities' sentiment.3.In order to solve the problem of partial entities without sentiment in the event,in the case that some entities' sentiment is known,we propose entities emotional inference methods based on equivalence relations and entities emotional inference methods based on the relationships between entities and verbs.We use the entities with sentiment to infer the sentiment of entities without sentiment and thus obtain more entities with sentiment.The research results of this paper can provide guidance for government decision-making and enterprise improvement products.The government understands public opinion through sentiment analysis so as to make decisions that suit the people's sentiment.The company continuously improves its products by analyzing users' evaluation to their products and mastering the pros and cons of products.