Multi-Granularity Sentiment Analysis Based On Text

The rapid development of science and technology makes the Internet and our lives more and more closely.With the sharply increasing number of netizens,much information about users' appears in the Internet.In order to enhance the relationship between the platform and users,the major websites allow a lot of various comments to be added into the platform.By this way,there has appeared a new method to get information.The appearance of information not only has changed the process of business in the website,but has a huge impact on user's mental model.Whereas,how to get these text information which includes characters,events,products,media and so on as quickly as they can,is still a question.So,analyzing these texts has been a vital issue.Text sentiment analysis is a part of web-text analysis issue.Text sentiment analysis,also called Opinion Mining or Review Mining,which is a process of processing,analysis and application with emotion.Text sentiment analysis is combined with network socialization media that plays a significant role in practical application.Text sentiment analysis has two major methods.The first one is based on dictionary,the other one is based on machine learning.The first method extracts emotional words from text as features.Then utilize dictionary to make a judgment about sentimental polarities.But,this method relays on the scale and quality of the dictionary.The second method utilizes machine learning by training algorithms to judge the sentimental polarities of words.However,this way needs artificial feature selection and prior knowledge and costs more time.To sum up,the main works are as follows:(1)Multi granularity sentiment analysis for text: The traditional methods,which are based on dictionary,just relay key words to make a judgment about sentimental polarities.It leads the appearance of error.So,this paper combines with the thinking of multiple granularities to judge the sentimental polarity of the text.These texts are divided into a granularity of word,of sentence and of document.By computing the multi granularity of information,this paper solves the problem of noise which is produced by the judgment of sentiment.(2)The change of sentiment with time: Any emotion is not invariable.With the event of time and subject,the text of network event also will change.This will cause text sentiment change too.So this paper proposed two states of sentiment: static sentiment and dynamic sentiment.According to different states of sentiment,we utilize different subject granularity and time dimension to compute and make a judgment about sentimental polarities.Then this paper shows some different maps such as the number of subjects of sentimental tendencies,number of sentimental tendencies at different time and so on.Comparing with traditional methods,the model proposed by this paper solves the dynamic sentimental analysis problem with the change of subject and time.