A Comparative Study of Lexical and Semantic Emoji Suggestion Systems

Mingrui 'Ray' Zhang, Alex Mariakakis, Jacob Burke, Jacob O. Wobbrock
We compared how lexical (word-based) and semantic (meaning-based) emoji suggestions mechanisms impact online conversations.


Emoji suggestion systems based on typed text have been proposed to encourage emoji usage and enrich text messaging; however, such systems’ actual effects on the chat experience are unknown. We built an Android keyboard with both lexical (word-based) and semantic (meaning-based) emoji suggestion capabilities and compared these in two different studies. To investigate the effect of emoji suggestion in online conversations, we conducted a laboratory text-messaging study with 24 participants and a 15-day longitudinal field deployment with 18 participants. We found that participants picked more semantic suggestions than lexical suggestions and perceived the semantic suggestions as more relevant to the message content. Our subjective data showed that although the suggestion mechanism did not affect the chatting experience significantly, different mechanisms could change the composing behavior of the users and facilitate their emoji-searching needs in different ways.