Toward nuanced sentiment analysis through multi-sense emoji embedding

Abstrak

This research investigates the role of emojis in sentiment analysis using a more comprehensive multi-sense skip-gram approach. Emojis, which can convey facial expressions, body movements, and intonations often challenging to express in text, enhance digital communication by enriching the meaning of messages. Previous studies have shown that emojis improve sentiment analysis, yet most focused solely on their positive and negative connotations. This study broadens the scope by incorporating positive, negative, and neutral sentiment contexts. In the experiments, emojis were embedded in text and converted into vector representations for further analysis. The classification of sentiment texts was performed using a bidirectional long short-term memory (Bi-LSTM) method enhanced with an attention layer. The experiments resulted in accuracy of 0.83, recall of 0.83, precision of 0.82, and F1-score of 0.82. Statistical tests confirmed the significance of these findings, indicating that the approach improves the accuracy of sentiment analysis involving emojis. Overall, the study demonstrates that the integration of text and emojis leads to a more nuanced and precise understanding of sentiment in sentences, confirming the effectiveness of this method.