Research On Aspect-based Opinion Mining Of Internet Reviews With Deep Learning

With the rapid development of network technology and the popularity of computer,the Internet has become an indispensable part of life,and people increasingly prefer to express opinions towards products,social problems and hot issues through the Internet.Internet reviews usually include multiple aspects of the objects,and these fine-grained information is also attracting more and more attention from consumers so that these information is worth further mining and analyzing.However,facing the increasing massive data of the Internet,it often takes a lot of time and effort for users to find the information that they have focused on.Therefore,we urgently need an automated method to implement aspect-specific opinion mining and analysis on mass reviews.The traditional opinion mining(OM)methods are not effective enough and lack efficient opnion summarization method.The work of this paper includes the following three aspects:First,this paper reviews the aspect-specific OM researches and methods.We summarize and analyze the existing works in the field from the perspectives of opinion aspect extraction,opinion expression extraction,OM evaluation metrics and common data resources.Second,this paper proposes a deep neural network with attention mechanism for the aspect-specific OM task inspired by the predecessors.Besides the semantic information,the input features of the model take into account the syntactic structure information of the input text.Experiments on multi-language data sets prove the efficiency of the model compared to baseline methods.Third,this paper proposes an optimization strategy for algorithm time complexity based on the traditional DBSCAN algorithm,and use the improved fast clustering algorithm to summarize the results of the aspect-specific OM.Clustering experiments on multiple data sets prove the efficiency and performance of the improved opinion summarization model.Fourth,we apply the aspect level mining model to the Chinese movie reviews data set to achieve an actor recommendation system.In this paper,a large number of original movie reviews are mined and summed up to form the raw opinion data,which are used to calculate the qualities of actors.According to the actor recommendation algorithm,we compare the new role with the the qualities of actors to achieve the actor recommendation's goal.This paper proposes a deep neural networks based aspect-specific OM model and a fast clustering method based on improved DBSCAN algorithm.Finally,the algorithms are integrated into the opinion mining system,and on this basis,we implement an actor recommendation system on Chinese movie review dataset,which can help users make better actor selection decisions for new film or new plays.