Improving Sentiment Detection in English Social Media Posts with Gated Recurrent Units and Decision Trees

Sentiment analysis plays a critical role in measuring social media consumption behavior, public opinion, and political sentiment on social media. This paper presents a Gated Recurrent Unit -Decision Trees (GRU-DT) hybrid model for sentiment classification of English social media text. GRUs are employed to learn temporal and contextual dependencies in sequential text data, managing long-term dependencies and vanishing gradient problems. The GRU features are fed into Decision Trees, which determine the sentiment as positive, negative, or neutral. Decision Trees provide better interpretability and overall classification accuracy. The model, implemented in Python, is shown to be 97.6% accurate— 19.49% compared to benchmark models such as MLP, CNN+Rule-Based, and Random Forest. The approach easily addresses problems such as slang and colloquial language typical in social media. Hybrid use of GRUs and Decision Trees for improved and interpretable sentiment analysis on social media text, which is often noisy and context-dependent.