Multidimensional Emotional Analysis Technology for Social Media Based on Viewpoint Extraction

With the global popularity of social media, how to effectively analyse the massive text data generated on these platforms to better understand users’ emotions and perspectives has become an important research direction. This study proposes a multidimensional sentiment analysis technique based on viewpoint extraction to overcome the shortcomings of traditional sentiment analysis methods in capturing emotional diversity and complexity. First, the study collects text data from various social media platforms, and after cleaning and preprocessing, constructs a sentiment analysis model that includes both serial and hybrid networks. In serial networks, a multi-layer architecture is adopted, including bidirectional encoders, convolutional neural networks, and bidirectional long short-term memory networks, to extract text features in an orderly manner. The hybrid network integrates the feature representations of different models and introduces a dual attention mechanism to enhance the ability to recognise evaluation objects and viewpoint holders. The results demonstrated that the proposed method exhibited enhanced accuracy, with improvements ranging from 1.51% to 0.96% in comparison to other serial or parallel models, and from 9.09% in comparison to other models. Introducing a dual attention mechanism significantly improves the accuracy of sentiment information extraction, with a performance improvement of about 5-6% compared to using only ordinary algorithms. This further substantiates the pivotal role of hierarchical feature extraction. Finally, the research findings provide a new perspective for social media sentiment analysis, which is expected to play an important role in practical applications such as marketing and public opinion monitoring. Further research will be conducted with the aim of expanding the data sample to enhance the model’s generalisation ability.