Automatic Emotion Recognition of Text Stories Based on Deep Learning

Due to the immense amounts of texts on the internet and the qualitative nature of human sentiment and the characteristics of machine learning and deep learning algorithms, they are potential candidates to be appli-ed in textual sentiment analysis. To compare the effectiveness of different algorithms, processed data using TF-IDF is input into different algorithms respectively, and the accuracy scores of the trials using the identical data-set are recorded for comparison. It turns out that the Extra Trees classifier and the Random Forest classi-fier performed the best among machine learning algorithms, suggesting the significance for reducing overfitting in this specific task given that the less overfitting-proof Decision Tree has performed worse. LSTM has a better ac-curacy score than CNN, though it is to be noted that the former runs significantly slower than the latter, indicating efficiency to be a potential topic to be considered.