Emojis are used frequently in social media and private conversations. They are significant means of communication that help us express emotions and describe objects visually. Previous studies have shown positive impact of emojis in human relations, memorization and user engagement with web content. Unicode version 6 includes 2923 emojis, which makes it hard to make full use of them without a recommender system. We formulate recommending emojis as a complex prediction problem based on its diverse usage as a word and as a sentiment marker.
People have individual usage of emojis, and different representation of emojis across different platforms also leads to different interpretations based on device.
Therefore, we introduce a recommender system that is able to suggest various emojis and apply personalization to increase the accuracy of the recommending process.
Exploring whether it is possible or not to extract knowledge from emoji datasets and using it to predict emoji usage, we implemented several baseline models and trained Long Short-Term Memory (LSTM) recurrent neural networks.
Master thesis paper
Master thesis presentation