Research On Aspect-level Sentiment Analysis Technology Oriented To Review Text

With the increasing popularity of Internet applications,people's lifestyles have undergone tremendous changes.Through the Internet,now people can buy goods,order meals,watch movies,etc.People's opinions on a certain product have formed a large number of comment texts.Obtaining consumers' emotional tendencies from these comments is of great significance for businesses or social organizations to make decisions.However,the current sentiment analysis method for comment text can only obtain an emotional tendency from a paragraph(or a sentence).For example,a certain review of a restaurant is marked as a good comment,but the consumer does not know whether the service is good or the food is good.The method does not accurately reflect the specific aspects that the consumer really cares about,but the aspect-level sentiment analysis can judge the emotional tendency of the text from a certain aspect.Based on this background,this paper conducts an aspect-level sentiment analysis study on the comment texts,which can extract the aspects that involve the consumer's concern and analyze the corresponding emotional tendencies.The research work of this paper is mainly as follows:(1)Aspect extraction,in order to obtain more aspects,this topic uses the part-of-speech sequence template to construct the candidate set,but the set contains a large number of semantically similar words,so an affinity propagation clustering algorithm based on cosine similarity is proposed in the paper.This algorithm clusters similar words into several categories,corresponding to various aspects in the comment texts.Compared with the traditional clustering algorithms,the algorithm performs better in clustering effect and time overhead.(2)Aspect-level sentiment analysis,in order to make full use of the aspect information in the comment text,this paper combines the aspect word embedding vector and the Long ShortTerm Memory model based on attention mechanism to form a new model.The corresponding emotional tendency prediction is carried out by this model,and the model has the highest accuracy in the comparison experiments.(3)System application,this topic uses the above algorithm and model to design and implement a recommendation system based on aspect-level sentiment analysis for product reviews.Starting from the system requirements,the system and the function of each module are designed separately,and the effectiveness of the system is tested.In this system,consumers can choose the most concerned aspects of a product,the system will calculate the recommended value of the product according to the choice of the consumers,then return the result to the consumer according to the value from high to low,so that it saves consumer's time that people read a large number of product reviews before making a purchase decision.