Advanced deep learning technology for emotion analysis on social media platform

When people want to express themselves online, microblogs are among the most popular platforms. Thus, emotion identification from microblogs has become a fascinating area of study for several reasons. Some problems with social media sentiment analysis include relying too much on human annotation, not capturing feature words having emotional hue well, and ignoring the long-distance semantic relationship of emotional characteristics. In order to accomplish the emotional evaluation of public opinion events on microblogs, this study presents a model for user emotion identification. The data collection and preparation of micro-blog comment language from public opinion events yields three distinct kinds of motivating text: "joy," "anger," and "sadness." Words describing emotions are then extracted by means of an algorithm that incorporates the Linear Discriminant Analysis (LDA) approach, an emotion dictionary, and human annotation. Using FastText, the inspiring text that was collected is transformed into a word vector. To complete the emotion classification, extract the text's main properties after collecting long-distance semantic data utilizing Siamese Capsule Networks. The test findings show that six machine learning models had an average F1 value increase of 4% and seven deep learning models had an average F1 value increase of 3%. When compared to existing machine learning as well as deep learning approaches, the proposed model outperforms them when it comes to determining the emotional state of social media users.