Research On Implicit Sentiment Analysis In News Comments

With the development of the Internet,the number of netizens has gradually increased,and the amount of network data has significantly growed.These data are based on the user's real experience so that they are valuable.Currently,sentiment analysis can be used to mine the emotions and opinions in data.However,sometimes people do not use emotional words directly,they choose to express what they really want to say "invisibly".Implicit sentiment analysis can solve such problems better.This kind of problem has a considerable proportion,and it has received less attention so far,therefore this dissertation focuses on implicit sentiment analysis and does research on various tasks among them.The main works of this dissertation are as followed:1.Construct an implicit sentiment analysis corpus,which is from news comments.Public corpus in implicit sentiment analysis don't match this dissertation,so it's better to build the corpus by ourselves.We introduce the construction process and labeling rules,and then analyze the theme and category distribution.2.Classify news comments' emotions are explicit or implicit.Judging whether the sentence contains explicit emotions or implicit emotions helps to use different methods to do better research.Therefore,this task is the basis of subsequent tasks.We propose a BCRNN model for this task,and compare the performance of machine learning models,integrated models,and deep learning models.3.Classify polarity of implicit sentiment.Sentiment polarity can greatly help us understand the users' attitude,and it is arguably the most important task.We propose an ECRNN model for this task,and combine various mechanisms to better identify sentiment polarity.4.Classify ways of expression of implicit sentiment.Due to the richness of natural language expression and the skewed data,this task is more difficult.We propose an AXCRNN model for this task,and compare the performance of different training set sizes and different methods of semi-supervised learning.5.Build a prototype system for implicit sentiment analysis in news comments.The system is easy to display and convenient to use.The finished system is not only easy to use and has a simple interface,but also has high practical value.