From the earliest texts and emails, users have sought to summarise what they wanted to say. Part of the impetus initially came from character limits on texts and ways to save bandwidth when using a slow internet connection in the dialup and pre-broadband days. Users would abbreviate common phrases, such as “laughing out loud” to “LOL” to express their amusement in a reply to a humorous message, for instance.
An alternative approach was to use punctuation marks to create a three-character icon, a smiley, that would represent a sentiment. For example, a colon followed by a hyphen and a left parenthesis sent in response to bad news would generally be interpreted at a sad face :-( whereas a happy face would use the parenthetical counterpart :-)
With increasing bandwidth and features on phones and devices came a need for more expressive alternatives to the simply smileys and abbreviations and a who alternative character set was devised to represent a wide range of facial expressions, hand gestures, objects, and activities. These tiny images work well with the feature-rich devices and greater bandwidth of 4G and 5G smart phones and other devices. They allow proficient users to express a wide range of emotions in a succinct way in their messages and even represent in a neat way complex ideas.
Of course, with our increasingly busy lives we are always looking for tools with which to shorten the time we need to produce even abbreviated messages. With text messages, autocorrect, autocomplete, and predictive text apps, generally work well to predict the end of a word or even the next word or words one is likely to need to type in a message. For instance, when arranging to meet a friend, one might finish a message by starting to type “see” and the device will predict the next words as “you soon”. But, how might this work with emoji?
Research in the International Journal of Business Intelligence and Data Mining, reveals a new approach to predicting which emoji a user is likely to use after a particular piece of text or other emoji. This new approach improves on earlier methods of emoji prediction by overcoming the problem that different users can have very disparate styles of emoji usage.
Vandita Grover and Hema Banati of the University of Delhi, India, have developed EmoRile to predict which emoji a user will use in a given conversation. The system creates a user profile based on past emoji use and tests show it to work at least as well as other emoji prediction tools. The EmoRile approach comes into its own when there is a much larger number of emoji to choose from for a given user profile.
The team points out that often users choose an emoji that might have the opposite meaning of the preceding text perhaps to express irony, sarcasm, or a witticism. This makes the emoji prediction task more difficult, so there is a need to develop a yet more nuanced algorithm to accurately predict emoji use in such messages. The presence of URLs, hashtags, and other context in a message might enhance emoji prediction accuracy.
Grover, V. and Banati, H. (2023) ‘EmoRile: a personalised emoji prediction scheme based on user profiling’, Int. J. Business Intelligence and Data Mining, Vol. 22, No. 4, pp.470–485.