Skip to content

In this project, we aim to achieve the task of predicting emojis from tweets. We aim to investigate the relationship between words and emojis.

License

Notifications You must be signed in to change notification settings

Karanjot-singh/Making-Emojis-More-Predictable

Repository files navigation

Making Emojis More Predictable

by Karan Abrol, Karanjot Singh and Pritish Wadhwa, Natural Language Processing (CSE546) under the guidance of Dr. Shad Akhtar from Indraprastha Institute of Information Technology, Delhi.

Introduction

The advent of social media platforms like WhatsApp, Facebook (Meta) and Twitter, etc. has changed natural language conversations forever. Emojis are small ideograms depicting objects, people, and scenes (Cappallo et al., 2015). Emojis are used to complement short text messages with a visual enhancement and have become a de-facto standard for online communication. Our aim is to predict a single emoji that appears in the input tweets.

In this project, we aim to achieve the task of predicting emojis from tweets. We aim to investigate the relationship between words and emojis.

Project Pipeline Summary

We started off by collecting the data. The data was then thoroughly studied and preprocessed. Key features were also extracted at this stage. Due to computational restrictions, a subset of data was taken which was further divided into training, test- ing and validation split, such that the distribution of any class in any two sets were same. After this, various machine learning and deep learning models were applied on the data set and the results were generated and analysed.

Deployment

Emoji Prediction Website

Screenshots

Prediction Website1 Prediction Website2

Dataset

The data we used consists of a list of tweets associated with a single emoji, indexed by 20 labels for each of the 20 emojis. 5,00,000 Tweets by users in the United States, from October 2015 to Jan 2018, were retrieved using the Twitter API. The script for scraping this dataset was made available by the SemEval 2018 challenge. Due to computational limitations we merged the test and trial data, and further divided that into training, trial and test data with a split of 70:10:20. We maintained the label ratios for each emoji across the three sets to best reflect how frequently they are used in real life.

Models

  • Machine Learning Models:

    • Logistic Regression
    • K-Nearest Neighbours
    • Stochastic Gradient Descent
    • Random Forest Classifier
    • Naive Bayes
    • Adaboost Classifier
    • Support Vector Machine
  • Deep Learning Models:

    • RNN
    • LSTM
    • BiLSTM

Contact

For further queries feel free to reach out to following contributors.
Karan Abrol ([email protected])
Karanjot Singh ([email protected])
Pritish Wadhwa ([email protected])

Final Report

Final Report 1
Final Report 2
Final Report 3
Final Report 4
Final Report 5
Final Report 6
Final Report 7

About

In this project, we aim to achieve the task of predicting emojis from tweets. We aim to investigate the relationship between words and emojis.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages