Question-answering Text Sentiment Analysis Based On Feature Fusion And Bi-Directional Attention Mechanism

With the boom in the Internet and e-commerce,sentiment analysis is widely used in academia and industry.However,as the number of e-commerce users and product reviews continues to grow,obtaining useful information becomes difficult.In addition,merchants use forms such as cashback for positive reviews to incentivise consumers to review,bringing about a large number of low-quality reviews.At the same time,there are even merchants who enhance the credibility of their shops and product sales by malicious means such as brushing up their reviews,further deteriorating the fairness of the platform and affecting the shopping experience of users.As a result,in recent years,mainstream e-commerce platforms including Amazon,Taobao,Jingdong and Net Ease Kaola have offered a new user-oriented approach to Q&A reviews.In this approach,platforms allow potential buyers to ask questions about a product and then invite quality users who have already purchased the product to answer.This approach makes up for the lack of traditional reviews and has become an important way for users to obtain information about products.Therefore,it is commercially important to study the text of question and answer reviews and to uncover the emotional messages they contain.The existing question-and-answer text sentiment analysis models do not sufficiently extract the semantic features of question-and-answer texts,and it is difficult to effectively deal with the problem of multiple meanings of words.This paper proposes a CBGRU-Bi ATT-BERT model based on feature fusion and bi-directional attention mechanism to address this shortcoming:(1)To tackle the challenges of word ambiguity and insufficient feature extraction in question-andanswer text,this paper utilizes BERT pre-training models to dynamically represent word embeddings,effectively mitigating the impact of word ambiguity on sentiment analysis in question-and-answer text.Furthermore,a hybrid neural network is constructed using parallel CNN and Bi GRU channels to extract text features.The CNN channel captures local features of question-and-answer text,while the Bi GRU channel combines forward GRU and backward GRU to primarily extract contextual features.This approach enables more comprehensive extraction of semantic features in question-andanswer text,resulting in more accurate classification results.(2)Due to the unique characteristics of question-and-answer text,the bidirectional attention mechanism is employed to simultaneously capture the interactive information between questions and answers.This mechanism enhances the understanding of the relationship between questions and answers and enables more accurate and comprehensive representations of text features.By introducing the bidirectional attention mechanism,the model becomes more flexible and accurate in capturing the sentiment interaction between questions and answers,thereby improving the performance and accuracy of sentiment analysis.Experimental results demonstrate the feasibility and superior performance of this method.(3)To present the results of question-and-answer text sentiment analysis in a more intuitive and clear manner,a prototype system based on feature fusion and bidirectional attention mechanism is designed and implemented,taking practical applications into account.The overall architecture design of the question-and-answer text sentiment analysis prototype system and the frontend page display of the prototype system are introduced based on system requirements analysis.Detailed explanations of the functionality of each module of the prototype system are provided.