Text Sentiment Analysis Of Social Media Reviews

With the rapid increase of social media user's reviews data,the sentiment analysis and opinion mining of user reviews texts have gradually attracted the attention of researchers.These reviews have a large amount of data and various styles,and manual processing alone is time-consuming and labor-intensive.Therefore,extracting sentiment information from social media reviews data,judging users' judgment attitudes toward products,services,etc.,and determining the focus and sentiment orientation of user reviews have very important research significance.Based on this,this paper uses deep learning model to mine the sentiment information of social media reviews data,and uses social media reviews data as the research object to focus on the two subtasks of text representation and sentiment classification model construction in the process of sentiment analysis.Firstly,in response to the needs of downstream tasks,this paper uses two text representation methods to digitally process the review text,named the distributed representation of the text and the graphical representation of the text.The distributed representation method is also called word embedding.This method usually inputs the words in the text into a pre-trained model for training,and converts them into continuous dense vectors.At present,distributed representation method is the most commonly used and most effective text representation method.It is based on the distributed assumption that words with the same or similar contexts have the same or similar meanings.The graph representation of the text is to obtain the part-of-speech information in the text and the grammatical relationship between words through syntactic analysis,using words as nodes and the relationship between words as edges to construct a text syntactic graph.Secondly,this paper proposed an interactive attention model(IAGCN)based on graph convolutional network based on the current fine-grained text sentiment analysis tasks that ignore the long-range dependence between words and the syntactic relationship.IAGCN fully considers the importance of the independent modeling of the target phrase and the interactive attention calculation between the target and the context for sentiment analysis.And we have used GCN to consider the syntactic information of the text,and calculate the influence of context words on the target phrase through syntactic distance attention,which improved the performance of the model.Finally,because people's understanding of language is inevitably affected by personal experience and cognitive level,it is strongly subjective.In other words,humans do not use language in isolation,and the use of language needs to consider its complex contextual information.This paper learns user embeddings and topic embeddings based on historical review documents of social media users and related topics,and reduces the impact of language subjectivity on sentiment classification performance by fusing complex contexts.Compared with ordinary text sentiment classification tasks,text classification combined with context can disambiguate complex languages.