A Study of Mongolian Emotion Classification Incorporating Emojis

Abstract

A Mongolian emotion classification algorithm incorporating emojis is proposed to address the problems of small Mongolian emotion classification corpus, poor classification results, and underutilization of emoji emotion features. Firstly, we extract Mongolian text data from the corpus, vectorize it using FastText algorithm and further learn the Mongolian text features. Secondly, emojis data are extracted from the corpus, vectorized and trained in GRU network to fully learn the emotion features of emoji. Then the attention mechanism is used to adjust the attention dynamics of text and emoji features in the model. Finally, the sentiment features of text and emoji are classified with softmax layer for sentiment classification. The experimental results show that the Mongolian sentiment classification algorithm with fused emojis outperforms FastText, Word2vec_BiLSTM and Glove _ BiLSTM sentiment classification algorithms in terms of precision, recall, F1 value and accuracy. The results show the effectiveness of the proposed method and provide a reference for Mongolian sentiment analysis and opinion prediction.