What does this Emoji Mean? A Vector Space Skip-Gram Model for Twitter Emojis
14916Authors
nan
Abstract
AbstractEmojis allow us to describe objects, situations and even feelings with small images, providing a visual and quick way to communicate. In this paper, we analyse emojis used in Twitter with distributional semantic models. We retrieve 10 millions tweets posted by USA users, and we build several skip gram word embedding models by mapping in the same vectorial space both words and emojis. We test our models with semantic similarity experiments, comparing the output of our models with human assessment. We also carry out an exhaustive qualitative evaluation, showing interesting results.
Keywords
Full Content
What does this Emoji Mean? A Vector Space Skip-Gram Model for Twitter Emojis Francesco Barbieri, Francesco Ronzano, Horacio Saggion Universitat Pompeu Fabra, Barcelona, Spain Tractament Autom`atic del Llenguatge Natural (TALN) name.surname@upf.edu Abstract Emojis allow us to describe objects, situations and even feelings with small images, providing a visual and quick way to communicate. In this paper, we analyse emojis used in Twitter with distributional semantic models. We retrieve 10 millions tweets posted by USA users, and we build several skip gram word embedding models by mapping in the same vectorial space both words and emojis. We test our models with semantic similarity experiments, comparing the output of our models with human assessment. We also carry out an exhaustive qualitative evaluation, showing interesting results. Keywords: Emojis, Social Networks, Embeddings 1. Introduction Writing and reading short texts is part of our daily life. There are very popular platforms on which we rely to com- municate our interests, opinions, emotions and daily ac- tivities using short texts, including Twitter, Instagram and WhatsApp. The users of such platforms are extremely di- verse with respect to their writing style and, more in gen- eral, with respect to the way they communicate with each other on these Social Media services. Yet, there is an as- pect they share: the use of emojis. Emojis1 are ideograms and smiles that can be considered the natural evolution of the emoticons (like :) and :D). There are several types of emojis, ranging from facial expressions to animals, ob- jects or places. Emojis use is increasing every day, rapidly changing the way in which we communicate in Social Net- works. Emojis are used to communicate simple things or feelings in a fresh, visual and condensed way. There are few studies that perform in depth analyses of emojis.2. We study emojis in the context of Twitter. Three tweets that include Twitter emojis3 are shown in Figure 1. Emojis can be used instead of words, like in the tweet of Kate Perry, where the emoji is used instead of America. Emojis are also exploited to denote the sentiment and real intents of a message, like in the Dwayne Johnson’s tweet. In his tweet, he makes a joke, suggesting to eat pancakes as they make you strong (denoting this with the arm emoji ), but at the end of the tweet he laughs (using ), suggesting that he is joking. In the tweet of Dawayne Johnson we can also see another emoji, a brown square . In this context it is used to sym- bolize a rock (as his stage name is The Rock), and close to the arm emoji, it would probably means “strong like a rock”, like himself. The use of emojis strongly depends 1Emojis comes originally from Japanese web texts. text2While very good analysis has been carried out in the con- of industries http://instagram-engineering. tumblr.com/post/117889701472/ emojineering-part-1-machine-learning-for-emoji 3https://blog.twitter.com/2014/ open-sourcing-twitter-emoji-for-everyone Figure 1: Examples of tweets that include emojis. on both the context and the community to which the same emoji is addressed. Even …
Content truncated to 500 words
Labeled Text
AbstractEmojis allow us to describe objects, situations and even feelings with small images, providing a visual and quick way to communicate. In this paper, we analyse emojis used in Twitter with distributional semantic models. We retrieve 10 millions tweets posted by USA users, and we build several skip gram word embedding models by mapping in the same vectorial space both words and emojis. We test our models with semantic similarity experiments, comparing the output of our models with human assessment. We also carry out an exhaustive qualitative evaluation, showing interesting results.
Publication Details
- Paper ID:
- 14916
- Year:
- 2016
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan