Design And Implementation Of A Sentiment-Stance Joint Learning Model Combined With Knowledge Graph

With the rapid development of the Internet and the rise of social media,more and more users tend to express their views and opinions on social topics and political events on social platforms.These opinions have a wide range of research value and application prospects in fields such as public opinion analysis and computational politics.Therefore,how to infer the potential stance of users on a certain topic in social media has become one of the hot research topics in recent years.Stance is defined as the expression of a reviewer’s attitude,opinion,and judgment about a target.Stance detection is an important branch of natural language processing,and is regarded as a subtask of text mining,aiming to judge users’ potential attitudes towards a given target based on the semantic information contained in the texts posted by users.Targets can be people,organizations,events,or policies.However,existing stance detection models do not take into account the distinction and connection between sentiment and stance,and lack full utilization of external knowledge.Benefiting from the idea of joint learning and the strategy of knowledge enhancement,this thesis conducts the following research work:(1)Propose a gated attention joint learning model for the target:the key motivation of this model is that it is not enough to judge the user’s stance simply by relying on the text,and the ordinary attention mechanism will be disturbed by irrelevant words.Therefore,this thesis implements a gated attention mechanism for target to detect stance.Compared with the ordinary attention mechanism,it pays more attention to words related to target and effectively avoids the interference of irrelevant words.In addition,based on the effectiveness of sentiment features and joint learning,this thesis implements a joint learning model with stance detection as the main task and sentiment analysis as the auxiliary task on the basis of the stance detection model based on gated attention.The effectiveness of gated attention and joint learning has been verified by comparative experiments.(2)Introduce external knowledge to joint learning model:this thesis proposes for the first time to introduce a variety of external knowledge(sentiment semantic knowledge and background knowledge)into the joint learning model.And in order to generate high-quality knowledge,a suitable knowledge graph will be constructed according to the specific corpus used.A variant of LSTM(Sentic LSTM)is used when integrating external knowledge.Experimental results show that the model with external knowledge has better performance,and the knowledge fusion method based on Sentic LSTM is better than simple vector splicing.(3)Build a visual demonstration system of social media stance detection:the system takes the Web terminal as the interactive entrance,uses the trained network model to automatically analyze sentiment and stance on a given target,and returns the results to the front-end interface.