Exploiting deep neural networks for tweet-based emoji prediction

Andrei Catalin Coman, Giacomo Zara, Yaroslav Nechaev, Gianni Barlacchi, Alessandro Moschitti

Research output: Contribution to journalConference article

Abstract

For many years now, emojis have been used in social networks and chat services in order to enrich written text with auxiliary graphical content, achieving a higher degree of empathy. In particular, given the wide use of this medium, emojis are now available in such a number and variety that every basic concept and mood is covered by at least one. For this reason, the connection between the emoji and its semantical meaning grows stronger. In this paper, we will be describing the work performed in order to develop a Machine Learning based tool that, given a tweet, predicts the most likely emoji associated with the text. The task resembles the one presented by Barbieri et al., [23], and is placed within the context of the International Workshop on Semantic Evaluation (SemEval) 2018. We designed a baseline with standard Support Vector Machines and another baseline based on fastText, which was provided as part of the Workshop. In addition, we implemented several models based on Neural Networks such as Bidirectional Long Short-Term Memory Recurrent Neural Networks and Convolutional Neural Networks. We found that the latter is the most effective since it outperformed all our models and ranks in the 6th position out of 47 total participants. Our work aims to illustrate the potential of simpler models, which, thanks to the fine-tuning of hyper-parameters, could achieve accuracy comparable to the more complex models of the challenge.

Original languageEnglish
Pages (from-to)116-128
Number of pages13
JournalCEUR Workshop Proceedings
Volume2244
Publication statusPublished - 1 Jan 2018
Event2nd Workshop on Natural Language for Artificial Intelligence, NL4AI 2018 - Trento, Italy
Duration: 22 Nov 201823 Nov 2018

Fingerprint

Neural networks
Recurrent neural networks
Support vector machines
Learning systems
Tuning
Semantics
Deep neural networks
Long short-term memory

Keywords

  • Bi-LSTM
  • CNN
  • Emoji
  • FastText
  • Machine Learning
  • SVM
  • Text Classification
  • Twitter

ASJC Scopus subject areas

  • Computer Science(all)

Cite this

Coman, A. C., Zara, G., Nechaev, Y., Barlacchi, G., & Moschitti, A. (2018). Exploiting deep neural networks for tweet-based emoji prediction. CEUR Workshop Proceedings, 2244, 116-128.

Exploiting deep neural networks for tweet-based emoji prediction. / Coman, Andrei Catalin; Zara, Giacomo; Nechaev, Yaroslav; Barlacchi, Gianni; Moschitti, Alessandro.

In: CEUR Workshop Proceedings, Vol. 2244, 01.01.2018, p. 116-128.

Research output: Contribution to journalConference article

Coman, AC, Zara, G, Nechaev, Y, Barlacchi, G & Moschitti, A 2018, 'Exploiting deep neural networks for tweet-based emoji prediction', CEUR Workshop Proceedings, vol. 2244, pp. 116-128.
Coman AC, Zara G, Nechaev Y, Barlacchi G, Moschitti A. Exploiting deep neural networks for tweet-based emoji prediction. CEUR Workshop Proceedings. 2018 Jan 1;2244:116-128.
Coman, Andrei Catalin ; Zara, Giacomo ; Nechaev, Yaroslav ; Barlacchi, Gianni ; Moschitti, Alessandro. / Exploiting deep neural networks for tweet-based emoji prediction. In: CEUR Workshop Proceedings. 2018 ; Vol. 2244. pp. 116-128.
@article{a6d96827827d4799b2e46f7d5b2acaf1,
title = "Exploiting deep neural networks for tweet-based emoji prediction",
abstract = "For many years now, emojis have been used in social networks and chat services in order to enrich written text with auxiliary graphical content, achieving a higher degree of empathy. In particular, given the wide use of this medium, emojis are now available in such a number and variety that every basic concept and mood is covered by at least one. For this reason, the connection between the emoji and its semantical meaning grows stronger. In this paper, we will be describing the work performed in order to develop a Machine Learning based tool that, given a tweet, predicts the most likely emoji associated with the text. The task resembles the one presented by Barbieri et al., [23], and is placed within the context of the International Workshop on Semantic Evaluation (SemEval) 2018. We designed a baseline with standard Support Vector Machines and another baseline based on fastText, which was provided as part of the Workshop. In addition, we implemented several models based on Neural Networks such as Bidirectional Long Short-Term Memory Recurrent Neural Networks and Convolutional Neural Networks. We found that the latter is the most effective since it outperformed all our models and ranks in the 6th position out of 47 total participants. Our work aims to illustrate the potential of simpler models, which, thanks to the fine-tuning of hyper-parameters, could achieve accuracy comparable to the more complex models of the challenge.",
keywords = "Bi-LSTM, CNN, Emoji, FastText, Machine Learning, SVM, Text Classification, Twitter",
author = "Coman, {Andrei Catalin} and Giacomo Zara and Yaroslav Nechaev and Gianni Barlacchi and Alessandro Moschitti",
year = "2018",
month = "1",
day = "1",
language = "English",
volume = "2244",
pages = "116--128",
journal = "CEUR Workshop Proceedings",
issn = "1613-0073",
publisher = "CEUR-WS",

}

TY - JOUR

T1 - Exploiting deep neural networks for tweet-based emoji prediction

AU - Coman, Andrei Catalin

AU - Zara, Giacomo

AU - Nechaev, Yaroslav

AU - Barlacchi, Gianni

AU - Moschitti, Alessandro

PY - 2018/1/1

Y1 - 2018/1/1

N2 - For many years now, emojis have been used in social networks and chat services in order to enrich written text with auxiliary graphical content, achieving a higher degree of empathy. In particular, given the wide use of this medium, emojis are now available in such a number and variety that every basic concept and mood is covered by at least one. For this reason, the connection between the emoji and its semantical meaning grows stronger. In this paper, we will be describing the work performed in order to develop a Machine Learning based tool that, given a tweet, predicts the most likely emoji associated with the text. The task resembles the one presented by Barbieri et al., [23], and is placed within the context of the International Workshop on Semantic Evaluation (SemEval) 2018. We designed a baseline with standard Support Vector Machines and another baseline based on fastText, which was provided as part of the Workshop. In addition, we implemented several models based on Neural Networks such as Bidirectional Long Short-Term Memory Recurrent Neural Networks and Convolutional Neural Networks. We found that the latter is the most effective since it outperformed all our models and ranks in the 6th position out of 47 total participants. Our work aims to illustrate the potential of simpler models, which, thanks to the fine-tuning of hyper-parameters, could achieve accuracy comparable to the more complex models of the challenge.

AB - For many years now, emojis have been used in social networks and chat services in order to enrich written text with auxiliary graphical content, achieving a higher degree of empathy. In particular, given the wide use of this medium, emojis are now available in such a number and variety that every basic concept and mood is covered by at least one. For this reason, the connection between the emoji and its semantical meaning grows stronger. In this paper, we will be describing the work performed in order to develop a Machine Learning based tool that, given a tweet, predicts the most likely emoji associated with the text. The task resembles the one presented by Barbieri et al., [23], and is placed within the context of the International Workshop on Semantic Evaluation (SemEval) 2018. We designed a baseline with standard Support Vector Machines and another baseline based on fastText, which was provided as part of the Workshop. In addition, we implemented several models based on Neural Networks such as Bidirectional Long Short-Term Memory Recurrent Neural Networks and Convolutional Neural Networks. We found that the latter is the most effective since it outperformed all our models and ranks in the 6th position out of 47 total participants. Our work aims to illustrate the potential of simpler models, which, thanks to the fine-tuning of hyper-parameters, could achieve accuracy comparable to the more complex models of the challenge.

KW - Bi-LSTM

KW - CNN

KW - Emoji

KW - FastText

KW - Machine Learning

KW - SVM

KW - Text Classification

KW - Twitter

UR - http://www.scopus.com/inward/record.url?scp=85057887474&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85057887474&partnerID=8YFLogxK

M3 - Conference article

VL - 2244

SP - 116

EP - 128

JO - CEUR Workshop Proceedings

JF - CEUR Workshop Proceedings

SN - 1613-0073

ER -