The Application Research Of Aspect Based Sentiment Analysis In Internet Review

With the wide application of Internet technology,Internet has become an important way to get data and share information and emotional communication in daily life.In order to help users to get more reassuring consumer experience and improve the services of businesses continuously,aspect based sentiment analysis technology came into being.It means the task needs to identify multiple attributes of the opinion target and their sentiment polarities of multiple attributes,such as the user's opinion of keyboard,memory and screen for laptop.The main research data comes from the SEMEVAL contest.Main job includes Opinion Target Expression Extraction,Aspect Category Identification and Category Based Sentiment Analysis.On the first task,we used the Conditional Random Field model(CRFs)and proposed two feature extraction schemes,respectively as the semantic features of comments and syntactic features of emotional dependency.Feature combination not only embodies the understanding of the semantics of the opinion target,but also explores the potential syntactic relations between the opinion target and the sentiment words.We got the top F1 score by combining two features which beats the best performance of SEMEVAL.Commenting on the second task,the Bidirectional Long Short-Term Memory(Bi-LSTM)is used,in which we integrated the last semantic vector from forward LSTM layer and backward LSTM layer directly.It not only simplifies the hyper parameters in the traditional model,and also promotes F1 score in non massive data.In the aspect based sentiment analysis,we put forward Bi-LSTM model based on attention mechanism.First,we got the word distribution in different categories using Labeled LDA as a supplementary feature of the model word vector,to get word distribution in different categories and make them more fitable to the given fields.Second,the given categories are represented by some words distribution.Finally,it made the attention mechanism functioned on the given category information and improved the accuracy of sentiment analysis.