Brand public opinion data analysis method based on deep learning

With the rapid development of Internet information technology and digital technology, various network platforms and media are showing a vigorous growth momentum. The powerful power of online public opinion has an immeasurable impact on brand awareness, consumption attitude, and decision-making of consumer groups. For brand owners, responding to sudden online public opinion has become an important issue and a new challenge in the current era. This article mainly focuses on Weibo comments, selects specific brand A, and tracks topic data related to public opinion events of brand A within a certain time range. The emotion dictionary is crucial in analyzing public opinion events using the Latent Dirichlet Allocation (LDA) topic model. This study aims to enhance the emotion dictionary by employing algorithms that leverage topic words and benchmark words, specifically in the context of Dalian University of Technology. The effectiveness of the improved emotion dictionary will be demonstrated through its integration with pre-trained word vectors, utilizing a Bidirectional Encoder Representations from Transformers (BERT) linear sentiment classification model. By combining these methods, the study seeks to provide more accurate sentiment analysis and deeper insights into public opinion. Finally, sentiment orientation analysis is conducted.