WCSE

CNN-BiLSTM with Attention Model for Emoji Recommendation

Abstract— Emojis are ideograms which are naturally combined with plain text to visually complement or emphasis the meaning of a message. In social communication, if plain text sentences are used in combination with different emojis, the meaning can be very different. Emoji recommendations are designed to solve the time-consuming issue of choosing an emoji among multiple possibilities on social platforms. The emoji recommendation task is defined as predicting the most likely emoji from a given text, as known as emoji prediction. Although emoji has been widely used on social platforms, the relationship between words and emojis is rarely researched by natural language processing (NLP). Also, the existing studies about emoji prediction are rarely based on Chinese corpus. For this purpose, we study the novel task of emoji prediction based on Chinese Weibo corpus. In this paper, we proposed a CNN-BiLSTM with attention model (CBLA) on the emoji prediction of Chinese social media. Our model is based on bidirectional long short-term memory (BiLSTM) layer with context-aware self-attention mechanism and convolutional layer (CNN). Experimental results show that our method achieved a good result and outperforms other prediction models.

Index Terms—emoji recommendation, long short-term memory, attention mechanism

Yan Wang, Yixian Di
School of Software Engineering, University of Science and Technology of China, CHINA
Viterbi School of Engineering, University of Southern California, UNITED STATES

[Download]

Cite: Yan Wang, Yixian Di, " CNN-BiLSTM with Attention Model for Emoji Recommendation " Proceedings of 2020 the 10th International Workshop on Computer Science and Engineering (WCSE 2020), pp. 62-69, Shanghai, China, 19-21 June, 2020.