Innovative Approaches to Sentiment Detection: Exploring Gated Recurrent Units and Decision Trees in Social Media

Sentiment detection is considered to be one of the most valuable tools in analyzing the target audience’s opinions and ongoing tendencies as well as in studying numerous customers feedbacks in the case of media platforms, for instance, Twitter. Nevertheless, the sentiment detection in social media data is not a trivial process because the information is often noisy, informal and context dependent. While assessing the sentiment, it is more appropriate to use techniques, which are much more complex and efficient such as, Bi-GRUs and MDTs are applied in this research. Bi-GRUs are used due to their insensitiveness to the position of the input in the sequence which makes them capable to comprehend past and future context that are important to understand the nature of the text from the social media platforms. To improve the stability of the model and its readability for a human expert, study introduce changes of the basic Decision Trees using an adaptive feature selection method involved with gradient boosting and implemented using python software. This is useful for the model to be able to learn feature importance in a dynamic model and thereby handle high dimensionality and minimize classification error. By reviewing the results shown by the hybrid model of Bi-GRUs and MDTs, it gets clear that the model possesses a high accuracy of a tremendous 99. 4%, compared to common baseline models of sentiment analysis in terms of precision, recall and F1-measure. This study demonstrates the applicability of integrating the state-of-the-art neural networks with richer decision tree improvements to the questions of sentiment analysis in social media, providing formidable instrument for various marketing, customer care, and other tasks.