Research On The Opinion Target Extraction Of News Comments

News comment opinion target extraction is an important technology for mining the views of netizens.This task aims to analyze the news comment text and extract the opinion target contained in it.The text of news comment can easily get the corresponding content of news comment,which has a high similarity with the comment.Based on the above characteristics,this thesis makes full use of the content of the news,and proposes an opinion target extraction model which integrates the news content.In order to solve the problem that the existing opinion target extraction model needs a large number of training samples but only less data samples,an opinion target extraction method based on external knowledge assisted dataset expansion is proposed.The thesis includes the following contents:1.The opinion target extraction based on news content.Based on the characteristics of news comment,this thesis first obtains candidate opinion target through focused named entity recognition,then constructs corresponding feature vectors by using the obtained candidate opinion target,and then learns the text and the comment text from the fusion of the candidate opinion targets' corresponding feature vectors and the vector input cyclic neural network of the comment text The characteristics of candidate opinion targets are input to conditional random fields for sequence annotation to extract the opinion target in the comments.The experimental results show that the integrates the news text information has a significant improvement compared with the not integrates the news text information,which shows that the method can better extract the opinion target in news comments after the fusion of text information.2.External knowledge assisted opinion target extraction of dataset expansion,aiming at the problem that the existing deep learning model needs a large number of training samples to fully train but the dataset samples are few.This thesis mainly proposes a method of opinion target extraction based on external knowledge assisted dataset expansion.First of all,a knowledge base of candidate opinion target is constructed by combining the candidate opinion target obtained from the text with the marked opinion target of relevant comments.Then,using this knowledge base,the method similar to distant supervision is used to automatically mark samples.In the marking of distant supervision method,two algorithms,direct matching and classifier matching,are proposed in this thesis.The experimental results show that the F1-value is significantly improved by using the method proposed in this thesis.The results show that the method based on external knowledge assisted dataset expansion is effective.3.The application of opinion target extraction.Based on the above research,this thesis implements the system,introduces the technologies and methods used in the implementation of the system,and verifies the effectiveness of the proposed model through the application in the real system The application results reflect the research value of this work.Through the above research and system implementation,this thesis makes full use of the characteristics of news comments,integrates the news text information,and proposes a deep learning model of integrating the news text information.To solve the problem that the existing deep learning model needs a large number of training samples in order to fully train but the dataset has a small number of samples,this thesis further proposes an opinion target extraction method that uses external knowledge to assist the expansion of the dataset to expand the dataset and improve the model effect.Finally,this thesis develops and implements the opinion target extraction system,and verifies the effectiveness of the work by using it in real scenarios.