Beyond Just Text: Semantic Emoji Similarity Modeling to Support Expressive Communication 👫📲😃
summary
Builds a semantic emoji model from 20.6 million tweets (≈228M tokens) using word2vec vectors to produce descriptor vectors for 845 emoji, shows the model outperforms a description-only (annotation) baseline in crowdsourced tests, and demonstrates a 2D projection and interactive demo plus cluster-based sampling for scalable keyboards. Claims the model improves contextual prediction and enables clustered keyboard prototypes; the trained vectors and data are available for reuse.
AI-written summary · about
chapters (11)
- 0:00Why study emoji
- 1:10Emoji entry challenge
- 2:11Emoji keyboard problems
- 5:18User study findings
- 7:48Semantic relationships
- 9:13Tweet-based data collection
- 10:38Word vector model
- 12:46Visualizing emoji space
- 13:48Future of emoji prediction
- 15:55Beyond emoji lists
- 16:59Resources and Q&A
related talks
20mExpressive Keyboards: Enriching Gesture-Typing on Mobile Devices
20mRepurposing Emoji for Personalised Communication: Why 🍕 means "I love you"
09mSymbolFinder: Brainstorming Diverse Symbols Using Local Semantic Networks
18mGoodbye Text, Hello Emoji: Mobile Communication on WeChat in China
02mHighlights - Images, the New Global Language
20mBridging the gap between AI and UI
23mDesigning for Accessibility
43mAI enhanced UX