Research On Sentiment Analysis In Social Networks

In recent years,with the development of Web 2.0 and the popularity of mobile portable devices,online social networks such as Facebook and We Chat have sprung up.Among them,the utilization rate of Microblog applications(such as Sina Weibo,Twitter,etc.)is far ahead.The development of online social networks makes the Internet infinitely close to the real world.On these platforms,users can manage their social relationships and social identities,publish content on various topics and get information from other users by the “following” connections.The resulting large amount of text data has attracted many scholars to study,text sentiment analysis has become a hot research topic in social network data analysis.Sentiment analysis has great application value in academic,social and commercial fields.However,most of the current social network sentiment analysis only analyzes the text and assumes that the text is independently and identically distributed,ignoring the impact of other information in the social network on sentiment analysis.For this problem,this paper takes real online social network data as the research object,combines the basic theories of sociology and psychology,and uses the basic nature of online social networks to study the sentiment analysis of texts in social networks from shallow to deep.The main research content of this paper includes the following four parts:Firstly,traditional microblog sentiment analysis methods assume that microblogs are independent and identically distributed and ignore the relationships between microblogs,so they usually have a low accuracy of sentiment analysis.To solve this problem,this paper proposes a sentiment analysis method based on user structure similarity and topic contexts.This method introduces user structure similarity to take the influence of common friends(seconddegree relationship)on sentiment analysis into consideration and formalizes it.We also introduce topic contexts to represent the semantic relationships between microblogs,and it is also formally represented.Based on the above,user structure similarity and topic contexts are unified as social contexts and combined with the microblog text feature classifier to form a new microblog sentiment analysis model.Extensive experimental and statistical analysis results show that the proposed method can achieve higher accuracy than traditional methods in microblog sentiment analysis.Secondly,traditional sentiment analysis methods cannot fully extract the heterogeneous relationships between microblogs,which affects the results of sentiment analysis.This paper proposes a sentiment analysis method based on weak dependency connections between microblogs.According to the theory of homogeneity,the community structure is a feature widely existing in social networks,and nodes in the same community often share the same properties.Based on this characteristic,this method assumes that weak dependency connections in social networks can affect microblog sentiment,and verifies the relevance between them via a statistical study.We establish a microblog relation graph according to user contexts and friend contexts,then we use community detection algorithms to extract the weak dependency connections between microblogs and formally represent them together with user contexts and friend contexts.Finally,text features of microblogs are combined to analyze microblog sentiment.Experiments on two real datasets show that our method can outperform state-of-art methods with a higher accuracy.Thirdly,methods exploring social contexts are usually based on traditional machine learning methods such as least squares.Moreover,they can only use social contexts during the model training stage,and cannot extract the deep features of social contexts and texts.This paper proposes a microblog sentiment analysis method based on social context representation learning.In this method,we establish a microblog relation graph based on the theory of sentimental consistency and emotional contagion and use a deep learning algorithm to map the nodes in the graph into a continuously distributed low-dimensional real vector space.This representation can mine the deep information of microblog relationships.A model based on the long short-term memory neural network is established.Social context vectors participate in different information calculations in this model,thus ensuring the maximization of information utilization of social context.In order to deal with the different weights of different words in sentiment analysis,we introduce an attention mechanism in the model.The experimental results on three datasets show that our proposed model can outperform state-of-the-art methods consistently and significantly in accuracy,precision and F1-score.Finally,we study one of the important applications of sentiment analysis in social networks,bursty event detection.To solve the problem of the low accuracy and efficiency of existing bursty event detection methods,this paper proposes a bursty event detection method based on sentiment co-occurrence graph and hashtag extraction.First,this method builds a sentiment cooccurrence graph that uses the Plutchik emotion wheel to define different sentiment type offline.Different from the traditional two-way and three-way classification methods,the sentiment cooccurrence graph can realize the fine-grained and unsupervised sentiment analysis of a microblog data stream so that the microblog data stream is divided into different sentiment data streams with a relatively small number of microblogs.Second,we perform burst detection on the obtained sentiment data streams and extract hashtags during the burst periods.At last,hashtags in the burst periods are segmented to obtain the candidate words of bursty events,the words which have a high correlation with the event candidate words in the sentiment data streams and the bursty event candidate words are selected as keywords for describing bursty events.This approach can detect bursty events online while analyzing the sentiment of microblogs.The experimental results show that our method can detect bursty events with high accuracy and in a short time.